{"thread":{"id":"66787","subject":"[PATCH 0/7] ci: some speedups for slow tests","startedAt":"2026-11-06T10:30:00Z","lastAt":"2026-11-12T08:19:22Z","messageCount":40,"participants":["Patrick Steinhardt","Junio C Hamano","Torsten Bögershausen","Johannes Sixt"],"isPatch":true,"patchVersion":1,"patchTotal":7},"messages":[{"id":"557648","messageId":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":null,"subject":"[PATCH 0/7] ci: some speedups for slow tests","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:00Z","receivedAt":"2026-11-06T10:30:00Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"Hi,\n\nI was staring at CI way more often recently, and that got me annoyed\nreal fast because it's just painstakingly slow. So I decided to have a\nlook and try do something about it, the result of which is this patch\nseries.\n\nThe series contains a couple of bits that are specific to GitLab, and a\ncouple of fixes that are generic. Some highlights:\n\n  - Installing the Git for Windows SDK speeds up from 3 minutes to 10\n    seconds by using tar(1) instead of Extract-Archive.\n\n  - t0027 speeds up from >10 minutes to 6 minutes on Windows by avoiding\n    the overhead of spawning redundant processes.\n\n  - t5551 and t5559 both speed up by a factor of 10 on macOS, going from\n    20 minutes each to 2 minutes each by using reftables in a repo with\n    100k refs. \"osx-meson\" and \"osx-clang\" thus go from 29 minutes to 18\n    minutes end-to-end.\n\n  - \"fedora-breaking-changes-meson\" speeds up by a factor of 2, going\n    from 35 minutes to 16 minutes. Turns out our SHA256 implementation\n    is quite slow with \"-O0\".\n\nUnfortunately, our Windows runners at GitLab are still atrociously slow\nby themselves. So the critical path only improves by about ~10 minutes,\ngoing from 60-70 minutes to 53 minutes end-to-end. But hey, at least the\nother jobs now give feedback a bit faster. And I guess on GitHub it may\nbe a different story because Windows runners are way better there and\nare typically not on the critical path.\n\nA test run can be found at [1].\n\nThanks!\n\nPatrick\n\n[1]: https://gitlab.com/gitlab-org/git/-/merge_requests/734\n\n---\nPatrick Steinhardt (7):\n      gitlab-ci: speed up installation of Git for Windows SDK\n      gitlab-ci: exclude intermediate outputs from Windows artifacts\n      ci: build optimized binaries with Meson\n      t0027: optimize how we reset repository state\n      t0027: cache expected output files\n      t1092: set up repositories from templates\n      t5551: use reftables for repositories with many refs\n\n .gitlab-ci.yml                           | 13 ++++++++++--\n ci/install-sdk.ps1                       | 26 ++++++++++++++++++------\n ci/run-build-and-tests.sh                |  1 +\n t/t0027-auto-crlf.sh                     | 33 +++++++++++++++++--------------\n t/t1092-sparse-checkout-compatibility.sh | 34 +++++++++++++++++++-------------\n t/t5551-http-fetch-smart.sh              |  6 ++++--\n 6 files changed, 74 insertions(+), 39 deletions(-)\n\n\n---\nbase-commit: 0ffa3e35c7c82ab8d0de226149fb03b3f7674d05\nchange-id: 20261106-pks-gitlab-ci-speedups-09c3841eb178\n\n\n"},{"id":"557649","messageId":"20261106-pks-gitlab-ci-speedups-v1-1-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:01Z","receivedAt":"2026-11-06T10:30:01Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"The \"build:mingw64\" job on GitLab CI downloads the minimal Git for\nWindows SDK on every run and then extracts it via `Expand-Archive`,\nwhich is expensive.\n\nImprove this by caching the SDK and by using tar(1) instead of\n`Expand-Archive`. Note that caching actually only gives us a win of a\ncouple seconds. The bigger win here is using tar(1), which speeds up\nextracting the archive from 3 minutes to about 10 seconds. Regardless of\nthat though, caching the SDK feels like a sensible thing to do as it\nremoves one step that may spuriously fail.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml     |  9 ++++++++-\n ci/install-sdk.ps1 | 26 ++++++++++++++++++++------\n 2 files changed, 28 insertions(+), 7 deletions(-)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 9b7698a0037..15419a954b3 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -127,9 +127,16 @@ build:mingw64:\n     - saas-windows-medium-amd64\n   variables:\n     NO_PERL: 1\n+  cache:\n+    key:\n+      files:\n+        - ci/install-sdk.ps1\n+        - ci/install-dependencies.ps1\n+    paths:\n+      - .dependencies\n   before_script:\n     - *windows_before_script\n-    - ./ci/install-sdk.ps1 -directory \"git-sdk\"\n+    - ./ci/install-sdk.ps1 -Directory \"git-sdk\"\n     - ./ci/install-dependencies.ps1 -Mingw\n   script:\n     # The minimal SDK's profile resets PATH.\ndiff --git a/ci/install-sdk.ps1 b/ci/install-sdk.ps1\nindex 66f24838a40..1dbcef04a18 100755\n--- a/ci/install-sdk.ps1\n+++ b/ci/install-sdk.ps1\n@@ -1,12 +1,26 @@\n param(\n-    [string]$directory='git-sdk',\n-    [string]$url='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip'\n+    [string]$Directory='git-sdk',\n+    [string]$URL='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip',\n+    [string]$DownloadDirectory='.dependencies'\n )\n \n-Invoke-WebRequest \"$url\" -OutFile git-sdk.zip\n-Expand-Archive -LiteralPath git-sdk.zip -DestinationPath \"$directory\"\n-Remove-Item -Path git-sdk.zip\n+$ErrorActionPreference = 'Stop'\n+$ProgressPreference = 'SilentlyContinue'\n+\n+New-Item -Path $DownloadDirectory -ItemType Directory -Force\n+\n+$archive = Join-Path $DownloadDirectory 'git-sdk.zip'\n+if (-not (Test-Path $archive)) {\n+    Invoke-WebRequest \"$URL\" -OutFile $archive -TimeoutSec 300\n+}\n+\n+New-Item -Path $Directory -ItemType Directory -Force\n+tar -xf $archive -C $Directory\n+if ($LASTEXITCODE -ne 0) {\n+    throw \"Failed to extract $archive with exit code $LASTEXITCODE\"\n+}\n \n New-Item -Path .git/info -ItemType Directory -Force\n New-Item -Path .git/info/exclude -ItemType File -Force\n-Add-Content -Path .git/info/exclude -Value \"/$directory\"\n+Add-Content -Path .git/info/exclude -Value \"/$Directory\"\n+Add-Content -Path .git/info/exclude -Value \"/$DownloadDirectory\"\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557650","messageId":"20261106-pks-gitlab-ci-speedups-v1-2-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 2/7] gitlab-ci: exclude intermediate outputs from Windows artifacts","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:02Z","receivedAt":"2026-11-06T10:30:02Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"As running tests is very slow on Windows we shard our CI tests across 10\nrunners. To avoid rebuilding Git on each of those runners we have a\npreliminary step which builds Git and then passes on the build artifacts\nto the test jobs.\n\nThis archive contains a ton of intermediate \"*.obj\" build artifacts that\naren't needed by the test jobs: the artifact contains ~2300 files, out\nof which ~1800 are \"*.obj\" files.\n\nExclude those files from the build artifact, which reduces the time it\ntakes to download and extract them from 34 seconds to 17 seconds.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml | 2 ++\n 1 file changed, 2 insertions(+)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 15419a954b3..036d2f3c7f2 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -194,6 +194,8 @@ build:msvc-meson:\n   artifacts:\n     paths:\n       - build\n+    exclude:\n+      - build/**/*.obj\n \n test:msvc-meson:\n   extends: .msvc-meson\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557651","messageId":"20261106-pks-gitlab-ci-speedups-v1-3-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 3/7] ci: build optimized binaries with Meson","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:03Z","receivedAt":"2026-11-06T10:30:03Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"When taking a look at our CI jobs one will quickly notice that across\nthe Linux-based jobs, the \"fedora-breaking-changes-meson\" job is running\nsignificantly longer than any of the other jobs. While most Linux jobs\nonly run around 15 minutes, that particular job runs roughly twice as\nlong.\n\nThe root cause of this is a combination of two different things:\n\n  - The job uses Meson to compile Git, and Meson by default uses the\n    \"debug\" build type that translates into \"-O0\". Our Makefile in\n    contrast to that uses \"-g -O2\" by default.\n\n  - The job enables breaking changes and thus runs with SHA256 by\n    default with the \"block\" backend, which benefits heavily from\n    compiler optimizations. For example, hashing a 500MB blob with\n    SHA256 takes 9.6 seconds with \"-O0\", whereas it takes only 1.5\n    seconds with \"-O2\". The SHA1 code path only slows down from 0.9\n    seconds to 2.1 seconds with \"-O0\".\n\nOf course, compiling without optimizations also has an impact on all the\nother jobs, but the Fedora job is by far the most impacted one.\n\nSpeed up executing the tests by using the \"debugoptimized\" build type\ninstead. This reduces the runtime of \"fedora-breaking-changes-meson\"\nfrom 2100 to 1000 seconds. Other Linux- and macOS-based jobs benefit as\nwell, but they only speed up by around ~10%.\n\nOn Windows though this is a bit more muddy, as the build step gets\nalmost two minutes slower, whereas tests only run ~10% faster. Combined\nthough the total run time is roughly the same before and after this\nchange.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml            | 2 +-\n ci/run-build-and-tests.sh | 1 +\n 2 files changed, 2 insertions(+), 1 deletion(-)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 036d2f3c7f2..df82cd3a5df 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -189,7 +189,7 @@ build:msvc-meson:\n   extends: .msvc-meson\n   stage: build\n   script:\n-    - meson setup build --vsenv -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n+    - meson setup build --vsenv --buildtype=debugoptimized -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n     - meson compile -C build\n   artifacts:\n     paths:\ndiff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh\nindex 68b91e31439..fd8136af342 100755\n--- a/ci/run-build-and-tests.sh\n+++ b/ci/run-build-and-tests.sh\n@@ -55,6 +55,7 @@ case \"$jobname\" in\n \t\t--fatal-meson-warnings \\\n \t\t--warnlevel 2 --werror \\\n \t\t--wrap-mode nofallback \\\n+\t\t--buildtype=debugoptimized \\\n \t\t-Dfuzzers=true \\\n \t\t-Dtest_output_directory=\"${TEST_OUTPUT_DIRECTORY:-$(pwd)/t}\" \\\n \t\t$MESONFLAGS\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557652","messageId":"20261106-pks-gitlab-ci-speedups-v1-4-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 4/7] t0027: optimize how we reset repository state","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:04Z","receivedAt":"2026-11-06T10:30:04Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In t0027 we verify our logic to convert CRLF for all kinds of\ncombinations of attributes and configuration, which ultimately results\nin 2600 test cases. This is of course on the heavier side, so on Windows\nthis takes more than 10 minutes to run.\n\nOne source of slowness is `checkout_files ()`, which gets called for\na lot of our test to set up the initial working directory. As part of\nthat we both remove the old files that we care about and then check them\nout again via git-checkout(1), but we do this one by one for every\nsingle file.\n\nThis is of course rather wasteful, as we have to spawn ten different\nprocesses every time the function is invoked. And on top of that, we\nalso have to rewrite the complete index in git-checkout(1) every time.\n\nSpeed this up by instead using a single invocation of rm(1) and\ngit-checkout(1). This reduces the runtime by about 20% on all platforms,\nand on Windows this reduces the runtime by ~2.5 minutes.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t0027-auto-crlf.sh | 16 +++++++++-------\n 1 file changed, 9 insertions(+), 7 deletions(-)\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex 49dbf09da77..eefb346d11f 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -303,13 +303,15 @@ checkout_files () {\n \t\tgit config core.autocrlf $crlf\n \t'\n \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n-\tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n-\tdo\n-\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n-\t\t\trm -f crlf_false_attr__$f.txt &&\n-\t\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- crlf_false_attr__$f.txt\n-\t\t'\n-\tdone\n+\ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n+\t\trm -f crlf_false_attr__LF.txt crlf_false_attr__CRLF.txt \\\n+\t\t\tcrlf_false_attr__LF_mix_CR.txt crlf_false_attr__CRLF_mix_LF.txt \\\n+\t\t\tcrlf_false_attr__LF_nul.txt &&\n+\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- \\\n+\t\t\tcrlf_false_attr__LF.txt crlf_false_attr__CRLF.txt \\\n+\t\t\tcrlf_false_attr__LF_mix_CR.txt crlf_false_attr__CRLF_mix_LF.txt \\\n+\t\t\tcrlf_false_attr__LF_nul.txt\n+\t'\n \n \ttest_expect_success \"ls-files --eol attr=$attr $ident aeol=$aeol core.autocrlf=$crlf core.eol=$ceol\" '\n \t\ttest_when_finished \"rm expect actual\" &&\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557653","messageId":"20261106-pks-gitlab-ci-speedups-v1-5-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 5/7] t0027: cache expected output files","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:05Z","receivedAt":"2026-11-06T10:30:05Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In the preceding commit we have optimized one aspect of t0027 already.\nAnother inefficiency is `compare_files ()` and `compare_ws_file ()`,\nwhich are both called more than a 1000 times.\n\nThese functions will repeatedly generate the same expected outcome even\nthough there's only a small handful of input files that the expected\noutcome is computed from. Recomputing those every time is of course\nquite pointless.\n\nImprove this by computing the expected outcome once, only, to reduce the\nnumber of spawned processes. This reduces the runtime by around 20-25%,\nand on Windows the runtime decreases from 480 to 380 seconds.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t0027-auto-crlf.sh | 17 +++++++++--------\n 1 file changed, 9 insertions(+), 8 deletions(-)\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex eefb346d11f..25ed6b33704 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -5,22 +5,23 @@ test_description='CRLF conversion all combinations'\n . ./test-lib.sh\n \n compare_files () {\n-\ttr '\\015\\000' QN <\"$1\" >\"$1\".expect &&\n+\t{ test -f \"$1\".expect || tr '\\015\\000' QN <\"$1\" >\"$1\".expect; } &&\n \ttr '\\015\\000' QN <\"$2\" | tr -d 'Z' >\"$2\".actual &&\n-\ttest_cmp \"$1\".expect \"$2\".actual &&\n-\trm \"$1\".expect \"$2\".actual\n+\ttest_cmp \"$1\".expect \"$2\".actual\n }\n \n compare_ws_file () {\n \tpfx=$1\n-\texp=$2.expect\n+\texp=$2.ws-expect\n \tact=$pfx.actual.$3\n-\ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$2\" |\n-\t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$exp\" &&\n+\t{\n+\t\ttest -f \"$exp\" ||\n+\t\ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$2\" |\n+\t\t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$exp\" ;\n+\t} &&\n \ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$3\" |\n \t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$act\" &&\n-\ttest_cmp \"$exp\" \"$act\" &&\n-\trm \"$exp\" \"$act\"\n+\ttest_cmp \"$exp\" \"$act\"\n }\n \n create_gitattributes () {\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557654","messageId":"20261106-pks-gitlab-ci-speedups-v1-6-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 6/7] t1092: set up repositories from templates","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:06Z","receivedAt":"2026-11-06T10:30:06Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In t1092 we verify the behaviour of sparse checkouts. We use three\ndifferent repositories to do so: one with a full checkout, one with a\nsparse checkout and one with a sparse index, and then verify that the\nresult is the same across all these configurations.\n\nThose repositories get set up via `init_repos ()`, which initializes\nthem from scratch every single time with a bunch of processes. The\nfunction is called more than a 100 times though, so the overhead adds up\nquite a bit. While the tests only require around 30 seconds on Linux,\nit takes more than 10 minutes on Windows.\n\nOptimize this by instead creating the repositories once, only. So\ninstead of recreating them from scratch, we can then just copy the\ntemplate repositories into place and thus save a bunch of processes.\nThis speeds up tests on Windows by about 3 minutes.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t1092-sparse-checkout-compatibility.sh | 34 +++++++++++++++++++-------------\n 1 file changed, 20 insertions(+), 14 deletions(-)\n\ndiff --git a/t/t1092-sparse-checkout-compatibility.sh b/t/t1092-sparse-checkout-compatibility.sh\nindex 05b54062b3b..5d81ec54c23 100755\n--- a/t/t1092-sparse-checkout-compatibility.sh\n+++ b/t/t1092-sparse-checkout-compatibility.sh\n@@ -141,28 +141,34 @@ test_expect_success 'setup' '\n \t)\n '\n \n+test_expect_success 'setup template repositories' '\n+\tcp -r initial-repo full-checkout.template &&\n+\tcp -r initial-repo sparse-checkout.template &&\n+\tcp -r initial-repo sparse-index.template &&\n+\n+\t# initialize sparse-checkout definitions\n+\tgit -C sparse-checkout.template sparse-checkout init --cone &&\n+\tgit -C sparse-checkout.template sparse-checkout set deep &&\n+\tgit -C sparse-index.template sparse-checkout init --cone --sparse-index &&\n+\ttest_cmp_config -C sparse-index.template true index.sparse &&\n+\tgit -C sparse-index.template sparse-checkout set deep &&\n+\n+\t# Disable this message to keep stderr the same.\n+\tgit -C sparse-index.template config advice.sparseIndexExpanded false\n+'\n+\n init_repos () {\n \trm -rf full-checkout sparse-checkout sparse-index &&\n \n \t# create repos in initial state\n-\tcp -r initial-repo full-checkout &&\n+\tcp -r full-checkout.template full-checkout &&\n \tgit -C full-checkout reset --hard &&\n \n-\tcp -r initial-repo sparse-checkout &&\n+\tcp -r sparse-checkout.template sparse-checkout &&\n \tgit -C sparse-checkout reset --hard &&\n \n-\tcp -r initial-repo sparse-index &&\n-\tgit -C sparse-index reset --hard &&\n-\n-\t# initialize sparse-checkout definitions\n-\tgit -C sparse-checkout sparse-checkout init --cone &&\n-\tgit -C sparse-checkout sparse-checkout set deep &&\n-\tgit -C sparse-index sparse-checkout init --cone --sparse-index &&\n-\ttest_cmp_config -C sparse-index true index.sparse &&\n-\tgit -C sparse-index sparse-checkout set deep &&\n-\n-\t# Disable this message to keep stderr the same.\n-\tgit -C sparse-index config advice.sparseIndexExpanded false\n+\tcp -r sparse-index.template sparse-index &&\n+\tgit -C sparse-index reset --hard\n }\n \n init_repos_as_submodules () {\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557655","messageId":"20261106-pks-gitlab-ci-speedups-v1-7-1fa0374c7287@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH 7/7] t5551: use reftables for repositories with many refs","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-06T10:30:07Z","receivedAt":"2026-11-06T10:30:07Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"Some of the tests in t5551 creates almost a 100,000 references and then\nverify that those can be fetched. Naturally, creating this many refs is\nquite involved with the \"files\" backend, as it means that we have to\ncreate this many individual loose files. This is tolerable on Linux,\nwhere the test takes around 2 minutes in our CI. But on macOS, the same\ntest takes almost 20 minutes to complete. And to add insult to injury,\nwe execute these tests a second time via t5559.\n\nThe system under test here is the HTTP negotiation, so we don't actually\ncare about whether or not the ref backend handles this alright. So adapt\nthe test to instead create repositories with the reftable backend, which\nis significantly faster in this case as we only need to write a single\ntable.\n\nAnd sure enough, this reduces the runtime of this test on macOS by a\nfactor of 10, going from 20 minutes to 2 minutes. On Linux, the tests at\nleast speed up by a factor of two.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t5551-http-fetch-smart.sh | 6 ++++--\n 1 file changed, 4 insertions(+), 2 deletions(-)\n\ndiff --git a/t/t5551-http-fetch-smart.sh b/t/t5551-http-fetch-smart.sh\nindex 130053c57aa..34945059531 100755\n--- a/t/t5551-http-fetch-smart.sh\n+++ b/t/t5551-http-fetch-smart.sh\n@@ -412,7 +412,8 @@ create_tags () {\n }\n \n test_expect_success 'create 2,000 tags in the repo' '\n-\tgit init --bare \"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n+\tgit init --bare --ref-format=reftable \\\n+\t\t\"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n \t(\n \t\tcd \"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n \t\tcreate_tags 1 2000\n@@ -421,7 +422,8 @@ test_expect_success 'create 2,000 tags in the repo' '\n \n test_expect_success CMDLINE_LIMIT \\\n \t'clone the 2,000 tag repo to check OS command line overflow' '\n-\trun_with_limited_cmdline git clone $HTTPD_URL/smart/many-tags.git too-many-refs &&\n+\trun_with_limited_cmdline git clone --ref-format=reftable \\\n+\t\t$HTTPD_URL/smart/many-tags.git too-many-refs &&\n \t(\n \t\tcd too-many-refs &&\n \t\tgit for-each-ref refs/tags >actual &&\n\n-- \n2.56.0.500.g5d06dc02cd.dirty\n\n\n"},{"id":"557714","messageId":"xmqqa4nllvuu.fsf@gitster.g","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-1-1fa0374c7287@pks.im","subject":"Re: [PATCH 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-06T17:20:57Z","receivedAt":"2026-11-06T17:20:57Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n> The \"build:mingw64\" job on GitLab CI downloads the minimal Git for\n> Windows SDK on every run and then extracts it via `Expand-Archive`,\n> which is expensive.\n>\n> Improve this by caching the SDK and by using tar(1) instead of\n> `Expand-Archive`.\n> Note that caching actually only gives us a win of a\n> couple seconds. The bigger win here is using tar(1), which speeds up\n> extracting the archive from 3 minutes to about 10 seconds. Regardless of\n\nWow, is a tool to move the bits from one part of a rusted iron\nsurface to another part really so much more expensive compared to\nanother tool to move the same bits, to make a difference that is\nworth doing this?\n\n> that though, caching the SDK feels like a sensible thing to do as it\n> removes one step that may spuriously fail.\n\nOne thing that may make readers curious is if we need to worry about\nthis cached copy going stale.\n\n> Signed-off-by: Patrick Steinhardt <ps@pks.im>\n> ---\n>  .gitlab-ci.yml     |  9 ++++++++-\n>  ci/install-sdk.ps1 | 26 ++++++++++++++++++++------\n>  2 files changed, 28 insertions(+), 7 deletions(-)\n>\n> diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\n> index 9b7698a0037..15419a954b3 100644\n> --- a/.gitlab-ci.yml\n> +++ b/.gitlab-ci.yml\n> @@ -127,9 +127,16 @@ build:mingw64:\n>      - saas-windows-medium-amd64\n>    variables:\n>      NO_PERL: 1\n> +  cache:\n> +    key:\n> +      files:\n> +        - ci/install-sdk.ps1\n> +        - ci/install-dependencies.ps1\n> +    paths:\n> +      - .dependencies\n>    before_script:\n>      - *windows_before_script\n> -    - ./ci/install-sdk.ps1 -directory \"git-sdk\"\n> +    - ./ci/install-sdk.ps1 -Directory \"git-sdk\"\n\nIs this change essential to understand this patch, or is it an\nunrelated general clean-up to follow some style rules?  I notice\nthat ...\n\n>      - ./ci/install-dependencies.ps1 -Mingw\n>    script:\n>      # The minimal SDK's profile resets PATH.\n> diff --git a/ci/install-sdk.ps1 b/ci/install-sdk.ps1\n> index 66f24838a40..1dbcef04a18 100755\n> --- a/ci/install-sdk.ps1\n> +++ b/ci/install-sdk.ps1\n> @@ -1,12 +1,26 @@\n>  param(\n> -    [string]$directory='git-sdk',\n> -    [string]$url='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip'\n> +    [string]$Directory='git-sdk',\n> +    [string]$URL='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip',\n> +    [string]$DownloadDirectory='.dependencies'\n\n... we are making a matching change here to upcase the parameter\nnames here.\n\n\n"},{"id":"557715","messageId":"xmqq5wy9lvpk.fsf@gitster.g","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-3-1fa0374c7287@pks.im","subject":"Re: [PATCH 3/7] ci: build optimized binaries with Meson","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-06T17:24:07Z","receivedAt":"2026-11-06T17:24:07Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n> Speed up executing the tests by using the \"debugoptimized\" build type\n> instead. This reduces the runtime of \"fedora-breaking-changes-meson\"\n> from 2100 to 1000 seconds. Other Linux- and macOS-based jobs benefit as\n> well, but they only speed up by around ~10%.\n\nThat is an impressive improvement.\n\n> On Windows though this is a bit more muddy, as the build step gets\n> almost two minutes slower, whereas tests only run ~10% faster. Combined\n> though the total run time is roughly the same before and after this\n> change.\n>\n> Signed-off-by: Patrick Steinhardt <ps@pks.im>\n> ---\n>  .gitlab-ci.yml            | 2 +-\n>  ci/run-build-and-tests.sh | 1 +\n>  2 files changed, 2 insertions(+), 1 deletion(-)\n>\n> diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\n> index 036d2f3c7f2..df82cd3a5df 100644\n> --- a/.gitlab-ci.yml\n> +++ b/.gitlab-ci.yml\n> @@ -189,7 +189,7 @@ build:msvc-meson:\n>    extends: .msvc-meson\n>    stage: build\n>    script:\n> -    - meson setup build --vsenv -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n> +    - meson setup build --vsenv --buildtype=debugoptimized -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n>      - meson compile -C build\n>    artifacts:\n>      paths:\n> diff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh\n> index 68b91e31439..fd8136af342 100755\n> --- a/ci/run-build-and-tests.sh\n> +++ b/ci/run-build-and-tests.sh\n> @@ -55,6 +55,7 @@ case \"$jobname\" in\n>  \t\t--fatal-meson-warnings \\\n>  \t\t--warnlevel 2 --werror \\\n>  \t\t--wrap-mode nofallback \\\n> +\t\t--buildtype=debugoptimized \\\n>  \t\t-Dfuzzers=true \\\n>  \t\t-Dtest_output_directory=\"${TEST_OUTPUT_DIRECTORY:-$(pwd)/t}\" \\\n>  \t\t$MESONFLAGS\n\n"},{"id":"557718","messageId":"xmqqv769kgqa.fsf@gitster.g","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-4-1fa0374c7287@pks.im","subject":"Re: [PATCH 4/7] t0027: optimize how we reset repository state","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-06T17:33:01Z","receivedAt":"2026-11-06T17:33:01Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n>  \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n> -\tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n> -\tdo\n> -\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n> -\t\t\trm -f crlf_false_attr__$f.txt &&\n> -\t\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- crlf_false_attr__$f.txt\n> -\t\t'\n> -\tdone\n> +\ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n> +\t\trm -f crlf_false_attr__LF.txt crlf_false_attr__CRLF.txt \\\n> +\t\t\tcrlf_false_attr__LF_mix_CR.txt crlf_false_attr__CRLF_mix_LF.txt \\\n> +\t\t\tcrlf_false_attr__LF_nul.txt &&\n> +\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- \\\n> +\t\t\tcrlf_false_attr__LF.txt crlf_false_attr__CRLF.txt \\\n> +\t\t\tcrlf_false_attr__LF_mix_CR.txt crlf_false_attr__CRLF_mix_LF.txt \\\n> +\t\t\tcrlf_false_attr__LF_nul.txt\n> +\t'\n\nHmph, don't we want avoid hand-expand the list of files even though\nwe may want to use one \"rm -f\" and another \"git checkout\" command\nto work on them?  IOW,\n\n\tpaths=\n\tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n\tdo\n\t\tpaths=${paths:+\"$paths \"}crlf_false_attr__$f.txt\n\tdone\n\ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n\t\trm -f $paths &&\n\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- $paths\n\t'\n\nor something?\n\n>  \n>  \ttest_expect_success \"ls-files --eol attr=$attr $ident aeol=$aeol core.autocrlf=$crlf core.eol=$ceol\" '\n>  \t\ttest_when_finished \"rm expect actual\" &&\n\n"},{"id":"557719","messageId":"xmqqqzgxkgf0.fsf@gitster.g","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"Re: [PATCH 0/7] ci: some speedups for slow tests","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-06T17:39:47Z","receivedAt":"2026-11-06T17:39:47Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n> I was staring at CI way more often recently, and that got me annoyed\n> real fast because it's just painstakingly slow. So I decided to have a\n> look and try do something about it, the result of which is this patch\n> series.\n\nThanks for these patches.  I didn't leave comments on later parts of\nthe series, but all of them were interesting read and fully made\nsense to me.\n\n"},{"id":"557845","messageId":"20261108164901.GA29213@tb-raspi4","threadId":"66787","inReplyTo":"xmqqv769kgqa.fsf@gitster.g","subject":"Re: [PATCH 4/7] t0027: optimize how we reset repository state","fromName":"Torsten Bögershausen","fromEmail":"tboegi@web.de","sentAt":"2026-11-08T16:49:01Z","receivedAt":"2026-11-08T16:49:01Z","isPatch":true,"sender":{"key":"tboegi@web.de","avatar":"https://avatars.githubusercontent.com/u/7138363?v=4"},"body":"On Fri, Nov 06, 2026 at 09:33:01AM -0800, Junio C Hamano wrote:\n> Hmph, don't we want avoid hand-expand the list of files even though\n> we may want to use one \"rm -f\" and another \"git checkout\" command\n> to work on them?  IOW,\n> \n> \tpaths=\n> \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n> \tdo\n> \t\tpaths=${paths:+\"$paths \"}crlf_false_attr__$f.txt\n> \tdone\n> \ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n> \t\trm -f $paths &&\n> \t\tgit ${ceol:+-c core.eol=$ceol} checkout -- $paths\n> \t'\n\n\"Wow, so it can go\" - a well working, but slow code is speed up so much.\n\nI used the following (before reading the suggestion from Junio)\nand the execution time here on an older MacOs dropped from 71 seconds to 54 seconds.\n\n    test_expect_success \"setup file checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n        files=\"\\\n           crlf_false_attr__LF.txt \\\n           crlf_false_attr__CRLF.txt \\\n           crlf_false_attr__LF_mix_CR.txt \\\n           crlf_false_attr__CRLF_mix_LF.txt \\\n           crlf_false_attr__LF_nul.txt\" &&\n        rm -fv $files &&\n        git ${ceol:+-c core.eol=$ceol} checkout -- $files\n    '\n\n"},{"id":"557875","messageId":"avFofjtc-1oXIPrB@pks.im","threadId":"66787","inReplyTo":"20261108164901.GA29213@tb-raspi4","subject":"Re: [PATCH 4/7] t0027: optimize how we reset repository state","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T06:30:22Z","receivedAt":"2026-11-09T06:30:22Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"On Sun, Nov 08, 2026 at 05:49:01PM +0100, Torsten Bögershausen wrote:\n> On Fri, Nov 06, 2026 at 09:33:01AM -0800, Junio C Hamano wrote:\n> > Hmph, don't we want avoid hand-expand the list of files even though\n> > we may want to use one \"rm -f\" and another \"git checkout\" command\n> > to work on them?  IOW,\n> > \n> > \tpaths=\n> > \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n> > \tdo\n> > \t\tpaths=${paths:+\"$paths \"}crlf_false_attr__$f.txt\n> > \tdone\n> > \ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n> > \t\trm -f $paths &&\n> > \t\tgit ${ceol:+-c core.eol=$ceol} checkout -- $paths\n> > \t'\n> \n> \"Wow, so it can go\" - a well working, but slow code is speed up so much.\n> \n> I used the following (before reading the suggestion from Junio)\n> and the execution time here on an older MacOs dropped from 71 seconds to 54 seconds.\n> \n>     test_expect_success \"setup file checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n>         files=\"\\\n>            crlf_false_attr__LF.txt \\\n>            crlf_false_attr__CRLF.txt \\\n>            crlf_false_attr__LF_mix_CR.txt \\\n>            crlf_false_attr__CRLF_mix_LF.txt \\\n>            crlf_false_attr__LF_nul.txt\" &&\n>         rm -fv $files &&\n>         git ${ceol:+-c core.eol=$ceol} checkout -- $files\n>     '\n\nHmm. I think I like Junio's alternative a bit better, but both are\ndefinitely an improvement over my take. Will add to the next version,\nthanks.\n\nPatrick\n\n"},{"id":"557876","messageId":"avFvC80wfg2nulG2@pks.im","threadId":"66787","inReplyTo":"xmqqa4nllvuu.fsf@gitster.g","subject":"Re: [PATCH 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T06:58:19Z","receivedAt":"2026-11-09T06:58:19Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"On Fri, Nov 06, 2026 at 09:20:57AM -0800, Junio C Hamano wrote:\n> Patrick Steinhardt <ps@pks.im> writes:\n> \n> > The \"build:mingw64\" job on GitLab CI downloads the minimal Git for\n> > Windows SDK on every run and then extracts it via `Expand-Archive`,\n> > which is expensive.\n> >\n> > Improve this by caching the SDK and by using tar(1) instead of\n> > `Expand-Archive`.\n> > Note that caching actually only gives us a win of a\n> > couple seconds. The bigger win here is using tar(1), which speeds up\n> > extracting the archive from 3 minutes to about 10 seconds. Regardless of\n> \n> Wow, is a tool to move the bits from one part of a rusted iron\n> surface to another part really so much more expensive compared to\n> another tool to move the same bits, to make a difference that is\n> worth doing this?\n\nI was surprised by this, as well.\n\n> > that though, caching the SDK feels like a sensible thing to do as it\n> > removes one step that may spuriously fail.\n> \n> One thing that may make readers curious is if we need to worry about\n> this cached copy going stale.\n\nYes, it may go stale, but I'm not even sure whether that really matters.\nI decided to just do this as-is and Cc Johannes to maybe facilitate his\ninput on this part. But arguably, we can just skip the cache altogether\nas it didn't really buy much anyway.\n\n> > diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\n> > index 9b7698a0037..15419a954b3 100644\n> > --- a/.gitlab-ci.yml\n> > +++ b/.gitlab-ci.yml\n> > @@ -127,9 +127,16 @@ build:mingw64:\n> >      - saas-windows-medium-amd64\n> >    variables:\n> >      NO_PERL: 1\n> > +  cache:\n> > +    key:\n> > +      files:\n> > +        - ci/install-sdk.ps1\n> > +        - ci/install-dependencies.ps1\n> > +    paths:\n> > +      - .dependencies\n> >    before_script:\n> >      - *windows_before_script\n> > -    - ./ci/install-sdk.ps1 -directory \"git-sdk\"\n> > +    - ./ci/install-sdk.ps1 -Directory \"git-sdk\"\n> \n> Is this change essential to understand this patch, or is it an\n> unrelated general clean-up to follow some style rules?  I notice\n> that ...\n> \n> >      - ./ci/install-dependencies.ps1 -Mingw\n> >    script:\n> >      # The minimal SDK's profile resets PATH.\n> > diff --git a/ci/install-sdk.ps1 b/ci/install-sdk.ps1\n> > index 66f24838a40..1dbcef04a18 100755\n> > --- a/ci/install-sdk.ps1\n> > +++ b/ci/install-sdk.ps1\n> > @@ -1,12 +1,26 @@\n> >  param(\n> > -    [string]$directory='git-sdk',\n> > -    [string]$url='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip'\n> > +    [string]$Directory='git-sdk',\n> > +    [string]$URL='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip',\n> > +    [string]$DownloadDirectory='.dependencies'\n> \n> ... we are making a matching change here to upcase the parameter\n> names here.\n\nWell, with the new DownloadDirectory parameter everything else would\nhave been inconsistent, so I made those changes while at it. I'll dropt\nhe cache in v2 though, and with that we can drop these changes, too.\n\nPatrick\n\n"},{"id":"557907","messageId":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH v2 0/7] ci: some speedups for slow tests","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:21Z","receivedAt":"2026-11-09T09:58:21Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"Hi,\n\nI was staring at CI way more often recently, and that got me annoyed\nreal fast because it's just painstakingly slow. So I decided to have a\nlook and try do something about it, the result of which is this patch\nseries.\n\nThe series contains a couple of bits that are specific to GitLab, and a\ncouple of fixes that are generic. Some highlights:\n\n  - Installing the Git for Windows SDK speeds up from 3 minutes to 10\n    seconds by using tar(1) instead of Extract-Archive.\n\n  - t0027 speeds up from >10 minutes to 6 minutes on Windows by avoiding\n    the overhead of spawning redundant processes.\n\n  - t5551 and t5559 both speed up by a factor of 10 on macOS, going from\n    20 minutes each to 2 minutes each by using reftables in a repo with\n    100k refs. \"osx-meson\" and \"osx-clang\" thus go from 29 minutes to 18\n    minutes end-to-end.\n\n  - \"fedora-breaking-changes-meson\" speeds up by a factor of 2, going\n    from 35 minutes to 16 minutes. Turns out our SHA256 implementation\n    is quite slow with \"-O0\".\n\nUnfortunately, our Windows runners at GitLab are still atrociously slow\nby themselves. So the critical path only improves by about ~10 minutes,\ngoing from 60-70 minutes to 53 minutes end-to-end. But hey, at least the\nother jobs now give feedback a bit faster. And I guess on GitHub it may\nbe a different story because Windows runners are way better there and\nare typically not on the critical path.\n\nA test run can be found at [1].\n\nChanges in v2:\n  - Drop the cache for the GfW SDK again.\n  - Drop repetitive paths in t0027.\n  - Link to v1: https://patch.msgid.link/20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im\n\nThanks!\n\nPatrick\n\n[1]: https://gitlab.com/gitlab-org/git/-/merge_requests/734\n\n---\nPatrick Steinhardt (7):\n      gitlab-ci: speed up installation of Git for Windows SDK\n      gitlab-ci: exclude intermediate outputs from Windows artifacts\n      ci: build optimized binaries with Meson\n      t0027: optimize how we reset repository state\n      t0027: cache expected output files\n      t1092: set up repositories from templates\n      t5551: use reftables for repositories with many refs\n\n .gitlab-ci.yml                           |  4 +++-\n ci/install-sdk.ps1                       |  6 +++++-\n ci/run-build-and-tests.sh                |  1 +\n t/t0027-auto-crlf.sh                     | 26 +++++++++++++-----------\n t/t1092-sparse-checkout-compatibility.sh | 34 +++++++++++++++++++-------------\n t/t5551-http-fetch-smart.sh              |  6 ++++--\n 6 files changed, 47 insertions(+), 30 deletions(-)\n\nRange-diff versus v1:\n\n1:  e25adb1b8c8 < -:  ----------- gitlab-ci: speed up installation of Git for Windows SDK\n-:  ----------- > 1:  582a9ab6468 gitlab-ci: speed up installation of Git for Windows SDK\n2:  116c8b8a716 = 2:  bff0a624127 gitlab-ci: exclude intermediate outputs from Windows artifacts\n3:  a035b8a902c = 3:  713e69bfb59 ci: build optimized binaries with Meson\n4:  1a1238f4be7 < -:  ----------- t0027: optimize how we reset repository state\n-:  ----------- > 4:  e421eb2a275 t0027: optimize how we reset repository state\n5:  c652f5b380a = 5:  16905b5be62 t0027: cache expected output files\n6:  9a90272c620 = 6:  43f355fbb02 t1092: set up repositories from templates\n7:  187128e7d5d = 7:  960716da178 t5551: use reftables for repositories with many refs\n\n---\nbase-commit: 0ffa3e35c7c82ab8d0de226149fb03b3f7674d05\nchange-id: 20261106-pks-gitlab-ci-speedups-09c3841eb178\n\n\n"},{"id":"557908","messageId":"20261109-pks-gitlab-ci-speedups-v2-1-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:22Z","receivedAt":"2026-11-09T09:58:22Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"The \"build:mingw64\" job on GitLab CI downloads the minimal Git for\nWindows SDK on every run and then extracts it via `Expand-Archive`,\nwhich is expensive.\n\nImprove this by using tar(1) instead of `Expand-Archive`. While at it,\nmake the script error on failures and suppress progress output.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n ci/install-sdk.ps1 | 6 +++++-\n 1 file changed, 5 insertions(+), 1 deletion(-)\n\ndiff --git a/ci/install-sdk.ps1 b/ci/install-sdk.ps1\nindex 66f24838a40..bb332c1c10f 100755\n--- a/ci/install-sdk.ps1\n+++ b/ci/install-sdk.ps1\n@@ -3,8 +3,12 @@ param(\n     [string]$url='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip'\n )\n \n+$ErrorActionPreference = 'Stop'\n+$ProgressPreference = 'SilentlyContinue'\n+\n Invoke-WebRequest \"$url\" -OutFile git-sdk.zip\n-Expand-Archive -LiteralPath git-sdk.zip -DestinationPath \"$directory\"\n+New-Item -Path $directory -ItemType Directory -Force\n+tar -xf git-sdk.zip -C $directory\n Remove-Item -Path git-sdk.zip\n \n New-Item -Path .git/info -ItemType Directory -Force\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557909","messageId":"20261109-pks-gitlab-ci-speedups-v2-2-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 2/7] gitlab-ci: exclude intermediate outputs from Windows artifacts","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:23Z","receivedAt":"2026-11-09T09:58:23Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"As running tests is very slow on Windows we shard our CI tests across 10\nrunners. To avoid rebuilding Git on each of those runners we have a\npreliminary step which builds Git and then passes on the build artifacts\nto the test jobs.\n\nThis archive contains a ton of intermediate \"*.obj\" build artifacts that\naren't needed by the test jobs: the artifact contains ~2300 files, out\nof which ~1800 are \"*.obj\" files.\n\nExclude those files from the build artifact, which reduces the time it\ntakes to download and extract them from 34 seconds to 17 seconds.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml | 2 ++\n 1 file changed, 2 insertions(+)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 9b7698a0037..3631393a8cc 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -187,6 +187,8 @@ build:msvc-meson:\n   artifacts:\n     paths:\n       - build\n+    exclude:\n+      - build/**/*.obj\n \n test:msvc-meson:\n   extends: .msvc-meson\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557910","messageId":"20261109-pks-gitlab-ci-speedups-v2-3-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 3/7] ci: build optimized binaries with Meson","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:24Z","receivedAt":"2026-11-09T09:58:24Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"When taking a look at our CI jobs one will quickly notice that across\nthe Linux-based jobs, the \"fedora-breaking-changes-meson\" job is running\nsignificantly longer than any of the other jobs. While most Linux jobs\nonly run around 15 minutes, that particular job runs roughly twice as\nlong.\n\nThe root cause of this is a combination of two different things:\n\n  - The job uses Meson to compile Git, and Meson by default uses the\n    \"debug\" build type that translates into \"-O0\". Our Makefile in\n    contrast to that uses \"-g -O2\" by default.\n\n  - The job enables breaking changes and thus runs with SHA256 by\n    default with the \"block\" backend, which benefits heavily from\n    compiler optimizations. For example, hashing a 500MB blob with\n    SHA256 takes 9.6 seconds with \"-O0\", whereas it takes only 1.5\n    seconds with \"-O2\". The SHA1 code path only slows down from 0.9\n    seconds to 2.1 seconds with \"-O0\".\n\nOf course, compiling without optimizations also has an impact on all the\nother jobs, but the Fedora job is by far the most impacted one.\n\nSpeed up executing the tests by using the \"debugoptimized\" build type\ninstead. This reduces the runtime of \"fedora-breaking-changes-meson\"\nfrom 2100 to 1000 seconds. Other Linux- and macOS-based jobs benefit as\nwell, but they only speed up by around ~10%.\n\nOn Windows though this is a bit more muddy, as the build step gets\nalmost two minutes slower, whereas tests only run ~10% faster. Combined\nthough the total run time is roughly the same before and after this\nchange.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml            | 2 +-\n ci/run-build-and-tests.sh | 1 +\n 2 files changed, 2 insertions(+), 1 deletion(-)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 3631393a8cc..449fed1b392 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -182,7 +182,7 @@ build:msvc-meson:\n   extends: .msvc-meson\n   stage: build\n   script:\n-    - meson setup build --vsenv -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n+    - meson setup build --vsenv --buildtype=debugoptimized -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n     - meson compile -C build\n   artifacts:\n     paths:\ndiff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh\nindex 68b91e31439..fd8136af342 100755\n--- a/ci/run-build-and-tests.sh\n+++ b/ci/run-build-and-tests.sh\n@@ -55,6 +55,7 @@ case \"$jobname\" in\n \t\t--fatal-meson-warnings \\\n \t\t--warnlevel 2 --werror \\\n \t\t--wrap-mode nofallback \\\n+\t\t--buildtype=debugoptimized \\\n \t\t-Dfuzzers=true \\\n \t\t-Dtest_output_directory=\"${TEST_OUTPUT_DIRECTORY:-$(pwd)/t}\" \\\n \t\t$MESONFLAGS\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557911","messageId":"20261109-pks-gitlab-ci-speedups-v2-4-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 4/7] t0027: optimize how we reset repository state","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:25Z","receivedAt":"2026-11-09T09:58:25Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In t0027 we verify our logic to convert CRLF for all kinds of\ncombinations of attributes and configuration, which ultimately results\nin 2600 test cases. This is of course on the heavier side, so on Windows\nthis takes more than 10 minutes to run.\n\nOne source of slowness is `checkout_files ()`, which gets called for\na lot of our test to set up the initial working directory. As part of\nthat we both remove the old files that we care about and then check them\nout again via git-checkout(1), but we do this one by one for every\nsingle file.\n\nThis is of course rather wasteful, as we have to spawn ten different\nprocesses every time the function is invoked. And on top of that, we\nalso have to rewrite the complete index in git-checkout(1) every time.\n\nSpeed this up by instead using a single invocation of rm(1) and\ngit-checkout(1). This reduces the runtime by about 20% on all platforms,\nand on Windows this reduces the runtime by ~2.5 minutes.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t0027-auto-crlf.sh | 9 +++++----\n 1 file changed, 5 insertions(+), 4 deletions(-)\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex 49dbf09da77..504a7ea54bb 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -305,11 +305,12 @@ checkout_files () {\n \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n \tdo\n-\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n-\t\t\trm -f crlf_false_attr__$f.txt &&\n-\t\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- crlf_false_attr__$f.txt\n-\t\t'\n+\t\tpaths=\"${paths:+\"$paths \"}crlf_false_attr__$f.txt\"\n \tdone\n+\ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n+\t\trm -f $paths &&\n+\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- $paths\n+\t'\n \n \ttest_expect_success \"ls-files --eol attr=$attr $ident aeol=$aeol core.autocrlf=$crlf core.eol=$ceol\" '\n \t\ttest_when_finished \"rm expect actual\" &&\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557912","messageId":"20261109-pks-gitlab-ci-speedups-v2-5-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 5/7] t0027: cache expected output files","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:26Z","receivedAt":"2026-11-09T09:58:26Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In the preceding commit we have optimized one aspect of t0027 already.\nAnother inefficiency is `compare_files ()` and `compare_ws_file ()`,\nwhich are both called more than a 1000 times.\n\nThese functions will repeatedly generate the same expected outcome even\nthough there's only a small handful of input files that the expected\noutcome is computed from. Recomputing those every time is of course\nquite pointless.\n\nImprove this by computing the expected outcome once, only, to reduce the\nnumber of spawned processes. This reduces the runtime by around 20-25%,\nand on Windows the runtime decreases from 480 to 380 seconds.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t0027-auto-crlf.sh | 17 +++++++++--------\n 1 file changed, 9 insertions(+), 8 deletions(-)\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex 504a7ea54bb..b021f0855a8 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -5,22 +5,23 @@ test_description='CRLF conversion all combinations'\n . ./test-lib.sh\n \n compare_files () {\n-\ttr '\\015\\000' QN <\"$1\" >\"$1\".expect &&\n+\t{ test -f \"$1\".expect || tr '\\015\\000' QN <\"$1\" >\"$1\".expect; } &&\n \ttr '\\015\\000' QN <\"$2\" | tr -d 'Z' >\"$2\".actual &&\n-\ttest_cmp \"$1\".expect \"$2\".actual &&\n-\trm \"$1\".expect \"$2\".actual\n+\ttest_cmp \"$1\".expect \"$2\".actual\n }\n \n compare_ws_file () {\n \tpfx=$1\n-\texp=$2.expect\n+\texp=$2.ws-expect\n \tact=$pfx.actual.$3\n-\ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$2\" |\n-\t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$exp\" &&\n+\t{\n+\t\ttest -f \"$exp\" ||\n+\t\ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$2\" |\n+\t\t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$exp\" ;\n+\t} &&\n \ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$3\" |\n \t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$act\" &&\n-\ttest_cmp \"$exp\" \"$act\" &&\n-\trm \"$exp\" \"$act\"\n+\ttest_cmp \"$exp\" \"$act\"\n }\n \n create_gitattributes () {\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557913","messageId":"20261109-pks-gitlab-ci-speedups-v2-6-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 6/7] t1092: set up repositories from templates","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:27Z","receivedAt":"2026-11-09T09:58:27Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In t1092 we verify the behaviour of sparse checkouts. We use three\ndifferent repositories to do so: one with a full checkout, one with a\nsparse checkout and one with a sparse index, and then verify that the\nresult is the same across all these configurations.\n\nThose repositories get set up via `init_repos ()`, which initializes\nthem from scratch every single time with a bunch of processes. The\nfunction is called more than a 100 times though, so the overhead adds up\nquite a bit. While the tests only require around 30 seconds on Linux,\nit takes more than 10 minutes on Windows.\n\nOptimize this by instead creating the repositories once, only. So\ninstead of recreating them from scratch, we can then just copy the\ntemplate repositories into place and thus save a bunch of processes.\nThis speeds up tests on Windows by about 3 minutes.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t1092-sparse-checkout-compatibility.sh | 34 +++++++++++++++++++-------------\n 1 file changed, 20 insertions(+), 14 deletions(-)\n\ndiff --git a/t/t1092-sparse-checkout-compatibility.sh b/t/t1092-sparse-checkout-compatibility.sh\nindex 05b54062b3b..5d81ec54c23 100755\n--- a/t/t1092-sparse-checkout-compatibility.sh\n+++ b/t/t1092-sparse-checkout-compatibility.sh\n@@ -141,28 +141,34 @@ test_expect_success 'setup' '\n \t)\n '\n \n+test_expect_success 'setup template repositories' '\n+\tcp -r initial-repo full-checkout.template &&\n+\tcp -r initial-repo sparse-checkout.template &&\n+\tcp -r initial-repo sparse-index.template &&\n+\n+\t# initialize sparse-checkout definitions\n+\tgit -C sparse-checkout.template sparse-checkout init --cone &&\n+\tgit -C sparse-checkout.template sparse-checkout set deep &&\n+\tgit -C sparse-index.template sparse-checkout init --cone --sparse-index &&\n+\ttest_cmp_config -C sparse-index.template true index.sparse &&\n+\tgit -C sparse-index.template sparse-checkout set deep &&\n+\n+\t# Disable this message to keep stderr the same.\n+\tgit -C sparse-index.template config advice.sparseIndexExpanded false\n+'\n+\n init_repos () {\n \trm -rf full-checkout sparse-checkout sparse-index &&\n \n \t# create repos in initial state\n-\tcp -r initial-repo full-checkout &&\n+\tcp -r full-checkout.template full-checkout &&\n \tgit -C full-checkout reset --hard &&\n \n-\tcp -r initial-repo sparse-checkout &&\n+\tcp -r sparse-checkout.template sparse-checkout &&\n \tgit -C sparse-checkout reset --hard &&\n \n-\tcp -r initial-repo sparse-index &&\n-\tgit -C sparse-index reset --hard &&\n-\n-\t# initialize sparse-checkout definitions\n-\tgit -C sparse-checkout sparse-checkout init --cone &&\n-\tgit -C sparse-checkout sparse-checkout set deep &&\n-\tgit -C sparse-index sparse-checkout init --cone --sparse-index &&\n-\ttest_cmp_config -C sparse-index true index.sparse &&\n-\tgit -C sparse-index sparse-checkout set deep &&\n-\n-\t# Disable this message to keep stderr the same.\n-\tgit -C sparse-index config advice.sparseIndexExpanded false\n+\tcp -r sparse-index.template sparse-index &&\n+\tgit -C sparse-index reset --hard\n }\n \n init_repos_as_submodules () {\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557914","messageId":"20261109-pks-gitlab-ci-speedups-v2-7-508832602d0b@pks.im","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"[PATCH v2 7/7] t5551: use reftables for repositories with many refs","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-09T09:58:28Z","receivedAt":"2026-11-09T09:58:28Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"Some of the tests in t5551 creates almost a 100,000 references and then\nverify that those can be fetched. Naturally, creating this many refs is\nquite involved with the \"files\" backend, as it means that we have to\ncreate this many individual loose files. This is tolerable on Linux,\nwhere the test takes around 2 minutes in our CI. But on macOS, the same\ntest takes almost 20 minutes to complete. And to add insult to injury,\nwe execute these tests a second time via t5559.\n\nThe system under test here is the HTTP negotiation, so we don't actually\ncare about whether or not the ref backend handles this alright. So adapt\nthe test to instead create repositories with the reftable backend, which\nis significantly faster in this case as we only need to write a single\ntable.\n\nAnd sure enough, this reduces the runtime of this test on macOS by a\nfactor of 10, going from 20 minutes to 2 minutes. On Linux, the tests at\nleast speed up by a factor of two.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t5551-http-fetch-smart.sh | 6 ++++--\n 1 file changed, 4 insertions(+), 2 deletions(-)\n\ndiff --git a/t/t5551-http-fetch-smart.sh b/t/t5551-http-fetch-smart.sh\nindex 130053c57aa..34945059531 100755\n--- a/t/t5551-http-fetch-smart.sh\n+++ b/t/t5551-http-fetch-smart.sh\n@@ -412,7 +412,8 @@ create_tags () {\n }\n \n test_expect_success 'create 2,000 tags in the repo' '\n-\tgit init --bare \"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n+\tgit init --bare --ref-format=reftable \\\n+\t\t\"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n \t(\n \t\tcd \"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n \t\tcreate_tags 1 2000\n@@ -421,7 +422,8 @@ test_expect_success 'create 2,000 tags in the repo' '\n \n test_expect_success CMDLINE_LIMIT \\\n \t'clone the 2,000 tag repo to check OS command line overflow' '\n-\trun_with_limited_cmdline git clone $HTTPD_URL/smart/many-tags.git too-many-refs &&\n+\trun_with_limited_cmdline git clone --ref-format=reftable \\\n+\t\t$HTTPD_URL/smart/many-tags.git too-many-refs &&\n \t(\n \t\tcd too-many-refs &&\n \t\tgit for-each-ref refs/tags >actual &&\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"557945","messageId":"20261109144651.GA14738@tb-raspi4","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-5-508832602d0b@pks.im","subject":"Re: [PATCH v2 5/7] t0027: cache expected output files","fromName":"Torsten Bögershausen","fromEmail":"tboegi@web.de","sentAt":"2026-11-09T14:46:51Z","receivedAt":"2026-11-09T14:46:51Z","isPatch":true,"sender":{"key":"tboegi@web.de","avatar":"https://avatars.githubusercontent.com/u/7138363?v=4"},"body":"On Mon, Nov 09, 2026 at 10:58:26AM +0100, Patrick Steinhardt wrote:\n> In the preceding commit we have optimized one aspect of t0027 already.\n> Another inefficiency is `compare_files ()` and `compare_ws_file ()`,\n> which are both called more than a 1000 times.\n> \n> These functions will repeatedly generate the same expected outcome even\n> though there's only a small handful of input files that the expected\n> outcome is computed from. Recomputing those every time is of course\n> quite pointless.\n> \n> Improve this by computing the expected outcome once, only, to reduce the\n> number of spawned processes. This reduces the runtime by around 20-25%,\n> and on Windows the runtime decreases from 480 to 380 seconds.\n> \n> Signed-off-by: Patrick Steinhardt <ps@pks.im>\n> ---\n>  t/t0027-auto-crlf.sh | 17 +++++++++--------\n>  1 file changed, 9 insertions(+), 8 deletions(-)\n> \n> diff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\n> index 504a7ea54bb..b021f0855a8 100755\n> --- a/t/t0027-auto-crlf.sh\n> +++ b/t/t0027-auto-crlf.sh\n> @@ -5,22 +5,23 @@ test_description='CRLF conversion all combinations'\n>  . ./test-lib.sh\n>  \n>  compare_files () {\n> -\ttr '\\015\\000' QN <\"$1\" >\"$1\".expect &&\n> +\t{ test -f \"$1\".expect || tr '\\015\\000' QN <\"$1\" >\"$1\".expect; } &&\n>  \ttr '\\015\\000' QN <\"$2\" | tr -d 'Z' >\"$2\".actual &&\n> -\ttest_cmp \"$1\".expect \"$2\".actual &&\n> -\trm \"$1\".expect \"$2\".actual\n> +\ttest_cmp \"$1\".expect \"$2\".actual\n>  }\n>  \n\nNice.\nIf you want to save even more CPU/SSD-cycles,\nyou can prepare the expect files like this:\n(No, I didn't do any benchmarks, nor look at ws_compare)\n\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex 49dbf09da7..3563336162 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -5,10 +5,9 @@ test_description='CRLF conversion all combinations'\n . ./test-lib.sh\n \n compare_files () {\n-\ttr '\\015\\000' QN <\"$1\" >\"$1\".expect &&\n \ttr '\\015\\000' QN <\"$2\" | tr -d 'Z' >\"$2\".actual &&\n-\ttest_cmp \"$1\".expect \"$2\".actual &&\n-\trm \"$1\".expect \"$2\".actual\n+\ttest_cmp \"expd/$1\".expect \"$2\".actual &&\n+\trm \"$2\".actual\n }\n \n compare_ws_file () {\n@@ -343,6 +342,25 @@ checkout_files () {\n \t\"\n }\n \n+test_expect_success 'create expect files' '\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\nLINEONE\\nLINETWO\\nLINETHREE\"     >LF &&\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\r\\nLINEONE\\r\\nLINETWO\\r\\nLINETHREE\" >CRLF &&\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\nLINEONE\\r\\nLINETWO\\nLINETHREE\"   >CRLF_mix_LF &&\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\nLINEONE\\nLINETWO\\rLINETHREE\"     >LF_mix_CR &&\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\r\\nLINEONE\\r\\nLINETWO\\rLINETHREE\"   >CRLF_mix_CR &&\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\r\\nLINEONEQ\\r\\nLINETWO\\r\\nLINETHREE\" | q_to_nul >CRLF_nul &&\n+\tprintf \"\\$Id: 0000000000000000000000000000000000000000 \\$\\nLINEONEQ\\nLINETWO\\nLINETHREE\" | q_to_nul >LF_nul &&\n+  mkdir expd &&\n+  echo \"* -text\" >expd/.gitattributes &&\n+\ttr \"\\015\\000\" QN <LF >expd/LF.expect &&\n+\ttr \"\\015\\000\" QN <CRLF >expd/CRLF.expect &&\n+\ttr \"\\015\\000\" QN <CRLF_mix_LF >expd/CRLF_mix_LF.expect &&\n+\ttr \"\\015\\000\" QN <LF_mix_CR >expd/LF_mix_CR.expect &&\n+\ttr \"\\015\\000\" QN <CRLF_mix_CR >expd/CRLF_mix_CR.expect &&\n+\ttr \"\\015\\000\" QN <CRLF_nul >expd/CRLF_nul.expect &&\n+\ttr \"\\015\\000\" QN <LF_nul >expd/LF_nul.expect\n+  '\n+\n # Test control characters\n # NUL SOH CR EOF==^Z\n test_expect_success 'ls-files --eol -o Text/Binary' '\n\n\n\n[snip]\n\n"},{"id":"557965","messageId":"xmqq8q320wv2.fsf@gitster.g","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-1-508832602d0b@pks.im","subject":"Re: [PATCH v2 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-09T16:53:53Z","receivedAt":"2026-11-09T16:53:53Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n> +tar -xf git-sdk.zip -C $directory\n\nWow.  You learn new things every day.\n\nThis is in ci/install-sdk.ps1 file.  Right now that script is\nreferred to only from .gitlab-ci.yml so as long as GitLab CI\nenvironment has tar that works this way we would be fine, but I\nthought that ci/* was designed to be portable across CI and other\nenvironments.  At least we would want a comment next to it that says\nwe assume BSD derived tar that can grok zip archives here, probably?\n\n\n"},{"id":"557966","messageId":"xmqq1p8u0wlg.fsf@gitster.g","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-4-508832602d0b@pks.im","subject":"Re: [PATCH v2 4/7] t0027: optimize how we reset repository state","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-09T16:59:39Z","receivedAt":"2026-11-09T16:59:39Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n> diff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\n> index 49dbf09da77..504a7ea54bb 100755\n> --- a/t/t0027-auto-crlf.sh\n> +++ b/t/t0027-auto-crlf.sh\n> @@ -305,11 +305,12 @@ checkout_files () {\n\nA micronit.\n\n>  \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n>  \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n>  \tdo\n> -\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n> -\t\t\trm -f crlf_false_attr__$f.txt &&\n> -\t\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- crlf_false_attr__$f.txt\n> -\t\t'\n> +\t\tpaths=\"${paths:+\"$paths \"}crlf_false_attr__$f.txt\"\n>  \tdone\n\nWe may know that nobody has touched variable \"$paths\" elsewhere in\nthis program, but I would prefer to have an explicit \"let's start\nwith an empty $paths\" assignment very close to this loop to help\nfuture readers.  That way they do not have to wonder and look all\nover to find out after what string we are appending these filenames.\n\n\n> +\ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n> +\t\trm -f $paths &&\n> +\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- $paths\n> +\t'\n>  \n>  \ttest_expect_success \"ls-files --eol attr=$attr $ident aeol=$aeol core.autocrlf=$crlf core.eol=$ceol\" '\n>  \t\ttest_when_finished \"rm expect actual\" &&\n\n"},{"id":"558023","messageId":"avK_Q_QDmdkl9r5s@pks.im","threadId":"66787","inReplyTo":"xmqq8q320wv2.fsf@gitster.g","subject":"Re: [PATCH v2 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-10T06:53:04Z","receivedAt":"2026-11-10T06:53:04Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"On Mon, Nov 09, 2026 at 08:53:53AM -0800, Junio C Hamano wrote:\n> Patrick Steinhardt <ps@pks.im> writes:\n> \n> > +tar -xf git-sdk.zip -C $directory\n> \n> Wow.  You learn new things every day.\n\nYeah, it's fun. Windows uses bsdtar, which is in turn based on\nlibarchive, and hence it also supports ZIP and 7zip files [1].\n\n> This is in ci/install-sdk.ps1 file.  Right now that script is\n> referred to only from .gitlab-ci.yml so as long as GitLab CI\n> environment has tar that works this way we would be fine, but I\n> thought that ci/* was designed to be portable across CI and other\n> environments.  At least we would want a comment next to it that says\n> we assume BSD derived tar that can grok zip archives here, probably?\n\nWell, bsdtar is nowadays distributed with Windows by default, so this\nshould be portable to other CI environments, too. So I don't think it\nreally warrants a comment as it should just work everwhere. That's\ncertainly something to point out in the commit message though.\n\nIn fact I wonder whether we should maybe adapt GitHub Workflows to use\nthat script, too. That was my intent back when introducing this script,\nand for a brief time we in fact did that. But we eventually figured that\nit was too slow in contrast to the\ngit-for-windows/setup-git-for-windows-sdk step, so we reverted that\nchange on GitHub again in 55d62306eea (GitHub ci(windows): speed up\ninitializing Git for Windows' minimal SDK again, 2024-12-17).\n\nBut now that we have found a way to remove the slowness we could unify\nboth systems again. I intentionally left that out of this patch series\nthough, as we can still do that as a subsequent step if we want to. In\nany case, I want Johannes' opinion on this topic before doing such a\nchange.\n\nPatrick\n\n[1]: https://learn.microsoft.com/en-us/windows/tar/\n\n"},{"id":"558024","messageId":"avK_V9z8gw20WS_M@pks.im","threadId":"66787","inReplyTo":"xmqq1p8u0wlg.fsf@gitster.g","subject":"Re: [PATCH v2 4/7] t0027: optimize how we reset repository state","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-10T06:53:11Z","receivedAt":"2026-11-10T06:53:11Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"On Mon, Nov 09, 2026 at 08:59:39AM -0800, Junio C Hamano wrote:\n> Patrick Steinhardt <ps@pks.im> writes:\n> \n> > diff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\n> > index 49dbf09da77..504a7ea54bb 100755\n> > --- a/t/t0027-auto-crlf.sh\n> > +++ b/t/t0027-auto-crlf.sh\n> > @@ -305,11 +305,12 @@ checkout_files () {\n> \n> A micronit.\n> \n> >  \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n> >  \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n> >  \tdo\n> > -\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n> > -\t\t\trm -f crlf_false_attr__$f.txt &&\n> > -\t\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- crlf_false_attr__$f.txt\n> > -\t\t'\n> > +\t\tpaths=\"${paths:+\"$paths \"}crlf_false_attr__$f.txt\"\n> >  \tdone\n> \n> We may know that nobody has touched variable \"$paths\" elsewhere in\n> this program, but I would prefer to have an explicit \"let's start\n> with an empty $paths\" assignment very close to this loop to help\n> future readers.  That way they do not have to wonder and look all\n> over to find out after what string we are appending these filenames.\n\nYeah, makes sense.\n\nPatrick\n\n"},{"id":"558025","messageId":"avK_XnoZ7Asakg25@pks.im","threadId":"66787","inReplyTo":"20261109144651.GA14738@tb-raspi4","subject":"Re: [PATCH v2 5/7] t0027: cache expected output files","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-10T06:53:18Z","receivedAt":"2026-11-10T06:53:18Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"On Mon, Nov 09, 2026 at 03:46:51PM +0100, Torsten Bögershausen wrote:\n> On Mon, Nov 09, 2026 at 10:58:26AM +0100, Patrick Steinhardt wrote:\n> > In the preceding commit we have optimized one aspect of t0027 already.\n> > Another inefficiency is `compare_files ()` and `compare_ws_file ()`,\n> > which are both called more than a 1000 times.\n> > \n> > These functions will repeatedly generate the same expected outcome even\n> > though there's only a small handful of input files that the expected\n> > outcome is computed from. Recomputing those every time is of course\n> > quite pointless.\n> > \n> > Improve this by computing the expected outcome once, only, to reduce the\n> > number of spawned processes. This reduces the runtime by around 20-25%,\n> > and on Windows the runtime decreases from 480 to 380 seconds.\n> > \n> > Signed-off-by: Patrick Steinhardt <ps@pks.im>\n> > ---\n> >  t/t0027-auto-crlf.sh | 17 +++++++++--------\n> >  1 file changed, 9 insertions(+), 8 deletions(-)\n> > \n> > diff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\n> > index 504a7ea54bb..b021f0855a8 100755\n> > --- a/t/t0027-auto-crlf.sh\n> > +++ b/t/t0027-auto-crlf.sh\n> > @@ -5,22 +5,23 @@ test_description='CRLF conversion all combinations'\n> >  . ./test-lib.sh\n> >  \n> >  compare_files () {\n> > -\ttr '\\015\\000' QN <\"$1\" >\"$1\".expect &&\n> > +\t{ test -f \"$1\".expect || tr '\\015\\000' QN <\"$1\" >\"$1\".expect; } &&\n> >  \ttr '\\015\\000' QN <\"$2\" | tr -d 'Z' >\"$2\".actual &&\n> > -\ttest_cmp \"$1\".expect \"$2\".actual &&\n> > -\trm \"$1\".expect \"$2\".actual\n> > +\ttest_cmp \"$1\".expect \"$2\".actual\n> >  }\n> >  \n> \n> Nice.\n> If you want to save even more CPU/SSD-cycles,\n> you can prepare the expect files like this:\n> (No, I didn't do any benchmarks, nor look at ws_compare)\n\nThanks! I benchmarked this and it doesn't seem to make much of a\ndifference. So I'd propose to stick with the simple change for now, if\nthat's alright with you.\n\nPatrick\n\n"},{"id":"558047","messageId":"20261110102907.GA30426@tb-raspi4","threadId":"66787","inReplyTo":"avK_XnoZ7Asakg25@pks.im","subject":"Re: [PATCH v2 5/7] t0027: cache expected output files","fromName":"Torsten Bögershausen","fromEmail":"tboegi@web.de","sentAt":"2026-11-10T10:29:07Z","receivedAt":"2026-11-10T10:29:07Z","isPatch":true,"sender":{"key":"tboegi@web.de","avatar":"https://avatars.githubusercontent.com/u/7138363?v=4"},"body":"\n> \n> Thanks! I benchmarked this and it doesn't seem to make much of a\n> difference. So I'd propose to stick with the simple change for now, if\n> that's alright with you.\n\nGood to hear that the benchmark was done.\nOK to stick your patch \"as is\"\n\n"},{"id":"558098","messageId":"xmqqwlqku0nv.fsf@gitster.g","threadId":"66787","inReplyTo":"avK_Q_QDmdkl9r5s@pks.im","subject":"Re: [PATCH v2 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Junio C Hamano","fromEmail":"gitster@pobox.com","sentAt":"2026-11-10T16:11:16Z","receivedAt":"2026-11-10T16:11:16Z","isPatch":true,"sender":{"key":"gitster@pobox.com","avatar":"https://avatars.githubusercontent.com/u/54884?v=4"},"body":"Patrick Steinhardt <ps@pks.im> writes:\n\n> On Mon, Nov 09, 2026 at 08:53:53AM -0800, Junio C Hamano wrote:\n>> Patrick Steinhardt <ps@pks.im> writes:\n>> \n>> > +tar -xf git-sdk.zip -C $directory\n>> \n>> Wow.  You learn new things every day.\n>\n> Yeah, it's fun. Windows uses bsdtar, which is in turn based on\n> libarchive, and hence it also supports ZIP and 7zip files [1].\n>\n>> This is in ci/install-sdk.ps1 file.  Right now that script is\n>> referred to only from .gitlab-ci.yml so as long as GitLab CI\n>> environment has tar that works this way we would be fine, but I\n>> thought that ci/* was designed to be portable across CI and other\n>> environments.  At least we would want a comment next to it that says\n>> we assume BSD derived tar that can grok zip archives here, probably?\n>\n> Well, bsdtar is nowadays distributed with Windows by default, so this\n> should be portable to other CI environments, too. So I don't think it\n> really warrants a comment as it should just work everwhere. That's\n> certainly something to point out in the commit message though.\n\nYeah, this installer is about Windows, so any current or future CI\nthat uses this script would end up running BSD tar, which means we\ncan assume readers know what the script is doing and why it works.\n\n> But now that we have found a way to remove the slowness we could unify\n> both systems again. I intentionally left that out of this patch series\n> though, as we can still do that as a subsequent step if we want to. In\n> any case, I want Johannes' opinion on this topic before doing such a\n> change.\n\nYes, indeed.  Thanks for these patches.\n\n"},{"id":"558171","messageId":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im","subject":"[PATCH v3 0/7] ci: some speedups for slow tests","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:51Z","receivedAt":"2026-11-11T10:47:51Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"Hi,\n\nI was staring at CI way more often recently, and that got me annoyed\nreal fast because it's just painstakingly slow. So I decided to have a\nlook and try do something about it, the result of which is this patch\nseries.\n\nThe series contains a couple of bits that are specific to GitLab, and a\ncouple of fixes that are generic. Some highlights:\n\n  - Installing the Git for Windows SDK speeds up from 3 minutes to 10\n    seconds by using tar(1) instead of Extract-Archive.\n\n  - t0027 speeds up from >10 minutes to 6 minutes on Windows by avoiding\n    the overhead of spawning redundant processes.\n\n  - t5551 and t5559 both speed up by a factor of 10 on macOS, going from\n    20 minutes each to 2 minutes each by using reftables in a repo with\n    100k refs. \"osx-meson\" and \"osx-clang\" thus go from 29 minutes to 18\n    minutes end-to-end.\n\n  - \"fedora-breaking-changes-meson\" speeds up by a factor of 2, going\n    from 35 minutes to 16 minutes. Turns out our SHA256 implementation\n    is quite slow with \"-O0\".\n\nUnfortunately, our Windows runners at GitLab are still atrociously slow\nby themselves. So the critical path only improves by about ~10 minutes,\ngoing from 60-70 minutes to 53 minutes end-to-end. But hey, at least the\nother jobs now give feedback a bit faster. And I guess on GitHub it may\nbe a different story because Windows runners are way better there and\nare typically not on the critical path.\n\nA test run can be found at [1].\n\nChanges in v3:\n  - Document why Windows' tar version can extract ZIP archives.\n  - Clear `paths` variable.\n  - Link to v2: https://patch.msgid.link/20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im\n\nChanges in v2:\n  - Drop the cache for the GfW SDK again.\n  - Drop repetitive paths in t0027.\n  - Link to v1: https://patch.msgid.link/20261106-pks-gitlab-ci-speedups-v1-0-1fa0374c7287@pks.im\n\nThanks!\n\nPatrick\n\n[1]: https://gitlab.com/gitlab-org/git/-/merge_requests/734\n\n---\nPatrick Steinhardt (7):\n      gitlab-ci: speed up installation of Git for Windows SDK\n      gitlab-ci: exclude intermediate outputs from Windows artifacts\n      ci: build optimized binaries with Meson\n      t0027: optimize how we reset repository state\n      t0027: cache expected output files\n      t1092: set up repositories from templates\n      t5551: use reftables for repositories with many refs\n\n .gitlab-ci.yml                           |  4 +++-\n ci/install-sdk.ps1                       |  6 +++++-\n ci/run-build-and-tests.sh                |  1 +\n t/t0027-auto-crlf.sh                     | 27 ++++++++++++++-----------\n t/t1092-sparse-checkout-compatibility.sh | 34 +++++++++++++++++++-------------\n t/t5551-http-fetch-smart.sh              |  6 ++++--\n 6 files changed, 48 insertions(+), 30 deletions(-)\n\nRange-diff versus v2:\n\n1:  41c3a0e8d7d ! 1:  8fabb336df3 gitlab-ci: speed up installation of Git for Windows SDK\n    @@ Commit message\n         Windows SDK on every run and then extracts it via `Expand-Archive`,\n         which is expensive.\n     \n    -    Improve this by using tar(1) instead of `Expand-Archive`. While at it,\n    -    make the script error on failures and suppress progress output.\n    +    Improve this by using tar(1) instead of `Expand-Archive`. Note that\n    +    we're using it to extract a ZIP archive, which may seem puzzling at\n    +    first. But Windows ships bsdtar by default [1], which also supports ZIP\n    +    and 7zip files.\n    +\n    +    While at it, make the script error on failures and suppress progress\n    +    output.\n    +\n    +    [1]: https://learn.microsoft.com/en-us/windows/tar/\n     \n         Signed-off-by: Patrick Steinhardt <ps@pks.im>\n     \n2:  439c12cb15d = 2:  4a4d1e6f2a8 gitlab-ci: exclude intermediate outputs from Windows artifacts\n3:  070bef6fecf = 3:  c23b7a7c58d ci: build optimized binaries with Meson\n4:  598a143fc24 ! 4:  942d70be750 t0027: optimize how we reset repository state\n    @@ Commit message\n     \n      ## t/t0027-auto-crlf.sh ##\n     @@ t/t0027-auto-crlf.sh: checkout_files () {\n    + \t\tgit config core.autocrlf $crlf\n    + \t'\n      \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n    ++\tpaths=\n      \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n      \tdo\n     -\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n5:  a325efbaddb = 5:  460cd0c88fd t0027: cache expected output files\n6:  a2d33441e50 = 6:  44c77717749 t1092: set up repositories from templates\n7:  bfc93fdfb4e = 7:  d64a61efec2 t5551: use reftables for repositories with many refs\n\n---\nbase-commit: 0ffa3e35c7c82ab8d0de226149fb03b3f7674d05\nchange-id: 20261106-pks-gitlab-ci-speedups-09c3841eb178\n\n\n"},{"id":"558172","messageId":"20261111-pks-gitlab-ci-speedups-v3-1-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 1/7] gitlab-ci: speed up installation of Git for Windows SDK","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:52Z","receivedAt":"2026-11-11T10:47:52Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"The \"build:mingw64\" job on GitLab CI downloads the minimal Git for\nWindows SDK on every run and then extracts it via `Expand-Archive`,\nwhich is expensive.\n\nImprove this by using tar(1) instead of `Expand-Archive`. Note that\nwe're using it to extract a ZIP archive, which may seem puzzling at\nfirst. But Windows ships bsdtar by default [1], which also supports ZIP\nand 7zip files.\n\nWhile at it, make the script error on failures and suppress progress\noutput.\n\n[1]: https://learn.microsoft.com/en-us/windows/tar/\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n ci/install-sdk.ps1 | 6 +++++-\n 1 file changed, 5 insertions(+), 1 deletion(-)\n\ndiff --git a/ci/install-sdk.ps1 b/ci/install-sdk.ps1\nindex 66f24838a40..bb332c1c10f 100755\n--- a/ci/install-sdk.ps1\n+++ b/ci/install-sdk.ps1\n@@ -3,8 +3,12 @@ param(\n     [string]$url='https://github.com/git-for-windows/git-sdk-64/releases/download/ci-artifacts/git-sdk-x86_64-minimal.zip'\n )\n \n+$ErrorActionPreference = 'Stop'\n+$ProgressPreference = 'SilentlyContinue'\n+\n Invoke-WebRequest \"$url\" -OutFile git-sdk.zip\n-Expand-Archive -LiteralPath git-sdk.zip -DestinationPath \"$directory\"\n+New-Item -Path $directory -ItemType Directory -Force\n+tar -xf git-sdk.zip -C $directory\n Remove-Item -Path git-sdk.zip\n \n New-Item -Path .git/info -ItemType Directory -Force\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558173","messageId":"20261111-pks-gitlab-ci-speedups-v3-2-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 2/7] gitlab-ci: exclude intermediate outputs from Windows artifacts","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:53Z","receivedAt":"2026-11-11T10:47:53Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"As running tests is very slow on Windows we shard our CI tests across 10\nrunners. To avoid rebuilding Git on each of those runners we have a\npreliminary step which builds Git and then passes on the build artifacts\nto the test jobs.\n\nThis archive contains a ton of intermediate \"*.obj\" build artifacts that\naren't needed by the test jobs: the artifact contains ~2300 files, out\nof which ~1800 are \"*.obj\" files.\n\nExclude those files from the build artifact, which reduces the time it\ntakes to download and extract them from 34 seconds to 17 seconds.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml | 2 ++\n 1 file changed, 2 insertions(+)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 9b7698a0037..3631393a8cc 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -187,6 +187,8 @@ build:msvc-meson:\n   artifacts:\n     paths:\n       - build\n+    exclude:\n+      - build/**/*.obj\n \n test:msvc-meson:\n   extends: .msvc-meson\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558174","messageId":"20261111-pks-gitlab-ci-speedups-v3-3-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 3/7] ci: build optimized binaries with Meson","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:54Z","receivedAt":"2026-11-11T10:47:54Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"When taking a look at our CI jobs one will quickly notice that across\nthe Linux-based jobs, the \"fedora-breaking-changes-meson\" job is running\nsignificantly longer than any of the other jobs. While most Linux jobs\nonly run around 15 minutes, that particular job runs roughly twice as\nlong.\n\nThe root cause of this is a combination of two different things:\n\n  - The job uses Meson to compile Git, and Meson by default uses the\n    \"debug\" build type that translates into \"-O0\". Our Makefile in\n    contrast to that uses \"-g -O2\" by default.\n\n  - The job enables breaking changes and thus runs with SHA256 by\n    default with the \"block\" backend, which benefits heavily from\n    compiler optimizations. For example, hashing a 500MB blob with\n    SHA256 takes 9.6 seconds with \"-O0\", whereas it takes only 1.5\n    seconds with \"-O2\". The SHA1 code path only slows down from 0.9\n    seconds to 2.1 seconds with \"-O0\".\n\nOf course, compiling without optimizations also has an impact on all the\nother jobs, but the Fedora job is by far the most impacted one.\n\nSpeed up executing the tests by using the \"debugoptimized\" build type\ninstead. This reduces the runtime of \"fedora-breaking-changes-meson\"\nfrom 2100 to 1000 seconds. Other Linux- and macOS-based jobs benefit as\nwell, but they only speed up by around ~10%.\n\nOn Windows though this is a bit more muddy, as the build step gets\nalmost two minutes slower, whereas tests only run ~10% faster. Combined\nthough the total run time is roughly the same before and after this\nchange.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n .gitlab-ci.yml            | 2 +-\n ci/run-build-and-tests.sh | 1 +\n 2 files changed, 2 insertions(+), 1 deletion(-)\n\ndiff --git a/.gitlab-ci.yml b/.gitlab-ci.yml\nindex 3631393a8cc..449fed1b392 100644\n--- a/.gitlab-ci.yml\n+++ b/.gitlab-ci.yml\n@@ -182,7 +182,7 @@ build:msvc-meson:\n   extends: .msvc-meson\n   stage: build\n   script:\n-    - meson setup build --vsenv -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n+    - meson setup build --vsenv --buildtype=debugoptimized -Dperl=disabled -Dbackend_max_links=1 -Dcredential_helpers=wincred -Dtest_output_directory=\"$TEST_OUTPUT_DIRECTORY\"\n     - meson compile -C build\n   artifacts:\n     paths:\ndiff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh\nindex 68b91e31439..fd8136af342 100755\n--- a/ci/run-build-and-tests.sh\n+++ b/ci/run-build-and-tests.sh\n@@ -55,6 +55,7 @@ case \"$jobname\" in\n \t\t--fatal-meson-warnings \\\n \t\t--warnlevel 2 --werror \\\n \t\t--wrap-mode nofallback \\\n+\t\t--buildtype=debugoptimized \\\n \t\t-Dfuzzers=true \\\n \t\t-Dtest_output_directory=\"${TEST_OUTPUT_DIRECTORY:-$(pwd)/t}\" \\\n \t\t$MESONFLAGS\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558175","messageId":"20261111-pks-gitlab-ci-speedups-v3-4-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 4/7] t0027: optimize how we reset repository state","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:55Z","receivedAt":"2026-11-11T10:47:55Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In t0027 we verify our logic to convert CRLF for all kinds of\ncombinations of attributes and configuration, which ultimately results\nin 2600 test cases. This is of course on the heavier side, so on Windows\nthis takes more than 10 minutes to run.\n\nOne source of slowness is `checkout_files ()`, which gets called for\na lot of our test to set up the initial working directory. As part of\nthat we both remove the old files that we care about and then check them\nout again via git-checkout(1), but we do this one by one for every\nsingle file.\n\nThis is of course rather wasteful, as we have to spawn ten different\nprocesses every time the function is invoked. And on top of that, we\nalso have to rewrite the complete index in git-checkout(1) every time.\n\nSpeed this up by instead using a single invocation of rm(1) and\ngit-checkout(1). This reduces the runtime by about 20% on all platforms,\nand on Windows this reduces the runtime by ~2.5 minutes.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t0027-auto-crlf.sh | 10 ++++++----\n 1 file changed, 6 insertions(+), 4 deletions(-)\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex 49dbf09da77..a3904b729b3 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -303,13 +303,15 @@ checkout_files () {\n \t\tgit config core.autocrlf $crlf\n \t'\n \tpfx=eol_${ceol}_crlf_${crlf}_attr_${attr}_ &&\n+\tpaths=\n \tfor f in LF CRLF LF_mix_CR CRLF_mix_LF LF_nul\n \tdo\n-\t\ttest_expect_success \"setup $f checkout ${ceol:+ with -c core.eol=$ceol}\"  '\n-\t\t\trm -f crlf_false_attr__$f.txt &&\n-\t\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- crlf_false_attr__$f.txt\n-\t\t'\n+\t\tpaths=\"${paths:+\"$paths \"}crlf_false_attr__$f.txt\"\n \tdone\n+\ttest_expect_success \"setup checkout ${ceol:+ with -c core.eol=$ceol}\" '\n+\t\trm -f $paths &&\n+\t\tgit ${ceol:+-c core.eol=$ceol} checkout -- $paths\n+\t'\n \n \ttest_expect_success \"ls-files --eol attr=$attr $ident aeol=$aeol core.autocrlf=$crlf core.eol=$ceol\" '\n \t\ttest_when_finished \"rm expect actual\" &&\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558176","messageId":"20261111-pks-gitlab-ci-speedups-v3-5-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 5/7] t0027: cache expected output files","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:56Z","receivedAt":"2026-11-11T10:47:56Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In the preceding commit we have optimized one aspect of t0027 already.\nAnother inefficiency is `compare_files ()` and `compare_ws_file ()`,\nwhich are both called more than a 1000 times.\n\nThese functions will repeatedly generate the same expected outcome even\nthough there's only a small handful of input files that the expected\noutcome is computed from. Recomputing those every time is of course\nquite pointless.\n\nImprove this by computing the expected outcome once, only, to reduce the\nnumber of spawned processes. This reduces the runtime by around 20-25%,\nand on Windows the runtime decreases from 480 to 380 seconds.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t0027-auto-crlf.sh | 17 +++++++++--------\n 1 file changed, 9 insertions(+), 8 deletions(-)\n\ndiff --git a/t/t0027-auto-crlf.sh b/t/t0027-auto-crlf.sh\nindex a3904b729b3..38ce06cbc1e 100755\n--- a/t/t0027-auto-crlf.sh\n+++ b/t/t0027-auto-crlf.sh\n@@ -5,22 +5,23 @@ test_description='CRLF conversion all combinations'\n . ./test-lib.sh\n \n compare_files () {\n-\ttr '\\015\\000' QN <\"$1\" >\"$1\".expect &&\n+\t{ test -f \"$1\".expect || tr '\\015\\000' QN <\"$1\" >\"$1\".expect; } &&\n \ttr '\\015\\000' QN <\"$2\" | tr -d 'Z' >\"$2\".actual &&\n-\ttest_cmp \"$1\".expect \"$2\".actual &&\n-\trm \"$1\".expect \"$2\".actual\n+\ttest_cmp \"$1\".expect \"$2\".actual\n }\n \n compare_ws_file () {\n \tpfx=$1\n-\texp=$2.expect\n+\texp=$2.ws-expect\n \tact=$pfx.actual.$3\n-\ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$2\" |\n-\t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$exp\" &&\n+\t{\n+\t\ttest -f \"$exp\" ||\n+\t\ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$2\" |\n+\t\t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$exp\" ;\n+\t} &&\n \ttr '\\015\\000abcdef0123456789' QN00000000000000000 <\"$3\" |\n \t\tsed -e \"s/0000*/$ZERO_OID/\" >\"$act\" &&\n-\ttest_cmp \"$exp\" \"$act\" &&\n-\trm \"$exp\" \"$act\"\n+\ttest_cmp \"$exp\" \"$act\"\n }\n \n create_gitattributes () {\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558177","messageId":"20261111-pks-gitlab-ci-speedups-v3-6-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 6/7] t1092: set up repositories from templates","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:57Z","receivedAt":"2026-11-11T10:47:57Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"In t1092 we verify the behaviour of sparse checkouts. We use three\ndifferent repositories to do so: one with a full checkout, one with a\nsparse checkout and one with a sparse index, and then verify that the\nresult is the same across all these configurations.\n\nThose repositories get set up via `init_repos ()`, which initializes\nthem from scratch every single time with a bunch of processes. The\nfunction is called more than a 100 times though, so the overhead adds up\nquite a bit. While the tests only require around 30 seconds on Linux,\nit takes more than 10 minutes on Windows.\n\nOptimize this by instead creating the repositories once, only. So\ninstead of recreating them from scratch, we can then just copy the\ntemplate repositories into place and thus save a bunch of processes.\nThis speeds up tests on Windows by about 3 minutes.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t1092-sparse-checkout-compatibility.sh | 34 +++++++++++++++++++-------------\n 1 file changed, 20 insertions(+), 14 deletions(-)\n\ndiff --git a/t/t1092-sparse-checkout-compatibility.sh b/t/t1092-sparse-checkout-compatibility.sh\nindex 05b54062b3b..5d81ec54c23 100755\n--- a/t/t1092-sparse-checkout-compatibility.sh\n+++ b/t/t1092-sparse-checkout-compatibility.sh\n@@ -141,28 +141,34 @@ test_expect_success 'setup' '\n \t)\n '\n \n+test_expect_success 'setup template repositories' '\n+\tcp -r initial-repo full-checkout.template &&\n+\tcp -r initial-repo sparse-checkout.template &&\n+\tcp -r initial-repo sparse-index.template &&\n+\n+\t# initialize sparse-checkout definitions\n+\tgit -C sparse-checkout.template sparse-checkout init --cone &&\n+\tgit -C sparse-checkout.template sparse-checkout set deep &&\n+\tgit -C sparse-index.template sparse-checkout init --cone --sparse-index &&\n+\ttest_cmp_config -C sparse-index.template true index.sparse &&\n+\tgit -C sparse-index.template sparse-checkout set deep &&\n+\n+\t# Disable this message to keep stderr the same.\n+\tgit -C sparse-index.template config advice.sparseIndexExpanded false\n+'\n+\n init_repos () {\n \trm -rf full-checkout sparse-checkout sparse-index &&\n \n \t# create repos in initial state\n-\tcp -r initial-repo full-checkout &&\n+\tcp -r full-checkout.template full-checkout &&\n \tgit -C full-checkout reset --hard &&\n \n-\tcp -r initial-repo sparse-checkout &&\n+\tcp -r sparse-checkout.template sparse-checkout &&\n \tgit -C sparse-checkout reset --hard &&\n \n-\tcp -r initial-repo sparse-index &&\n-\tgit -C sparse-index reset --hard &&\n-\n-\t# initialize sparse-checkout definitions\n-\tgit -C sparse-checkout sparse-checkout init --cone &&\n-\tgit -C sparse-checkout sparse-checkout set deep &&\n-\tgit -C sparse-index sparse-checkout init --cone --sparse-index &&\n-\ttest_cmp_config -C sparse-index true index.sparse &&\n-\tgit -C sparse-index sparse-checkout set deep &&\n-\n-\t# Disable this message to keep stderr the same.\n-\tgit -C sparse-index config advice.sparseIndexExpanded false\n+\tcp -r sparse-index.template sparse-index &&\n+\tgit -C sparse-index reset --hard\n }\n \n init_repos_as_submodules () {\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558178","messageId":"20261111-pks-gitlab-ci-speedups-v3-7-d1e34117ee2e@pks.im","threadId":"66787","inReplyTo":"20261111-pks-gitlab-ci-speedups-v3-0-d1e34117ee2e@pks.im","subject":"[PATCH v3 7/7] t5551: use reftables for repositories with many refs","fromName":"Patrick Steinhardt","fromEmail":"ps@pks.im","sentAt":"2026-11-11T10:47:58Z","receivedAt":"2026-11-11T10:47:58Z","isPatch":true,"sender":{"key":"ps@pks.im","avatar":"https://avatars.githubusercontent.com/u/4056630?v=4"},"body":"Some of the tests in t5551 creates almost a 100,000 references and then\nverify that those can be fetched. Naturally, creating this many refs is\nquite involved with the \"files\" backend, as it means that we have to\ncreate this many individual loose files. This is tolerable on Linux,\nwhere the test takes around 2 minutes in our CI. But on macOS, the same\ntest takes almost 20 minutes to complete. And to add insult to injury,\nwe execute these tests a second time via t5559.\n\nThe system under test here is the HTTP negotiation, so we don't actually\ncare about whether or not the ref backend handles this alright. So adapt\nthe test to instead create repositories with the reftable backend, which\nis significantly faster in this case as we only need to write a single\ntable.\n\nAnd sure enough, this reduces the runtime of this test on macOS by a\nfactor of 10, going from 20 minutes to 2 minutes. On Linux, the tests at\nleast speed up by a factor of two.\n\nSigned-off-by: Patrick Steinhardt <ps@pks.im>\n---\n t/t5551-http-fetch-smart.sh | 6 ++++--\n 1 file changed, 4 insertions(+), 2 deletions(-)\n\ndiff --git a/t/t5551-http-fetch-smart.sh b/t/t5551-http-fetch-smart.sh\nindex 130053c57aa..34945059531 100755\n--- a/t/t5551-http-fetch-smart.sh\n+++ b/t/t5551-http-fetch-smart.sh\n@@ -412,7 +412,8 @@ create_tags () {\n }\n \n test_expect_success 'create 2,000 tags in the repo' '\n-\tgit init --bare \"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n+\tgit init --bare --ref-format=reftable \\\n+\t\t\"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n \t(\n \t\tcd \"$HTTPD_DOCUMENT_ROOT_PATH/many-tags.git\" &&\n \t\tcreate_tags 1 2000\n@@ -421,7 +422,8 @@ test_expect_success 'create 2,000 tags in the repo' '\n \n test_expect_success CMDLINE_LIMIT \\\n \t'clone the 2,000 tag repo to check OS command line overflow' '\n-\trun_with_limited_cmdline git clone $HTTPD_URL/smart/many-tags.git too-many-refs &&\n+\trun_with_limited_cmdline git clone --ref-format=reftable \\\n+\t\t$HTTPD_URL/smart/many-tags.git too-many-refs &&\n \t(\n \t\tcd too-many-refs &&\n \t\tgit for-each-ref refs/tags >actual &&\n\n-- \n2.56.0.611.g1030f0b940.dirty\n\n\n"},{"id":"558280","messageId":"0182f66a-f602-4ec6-8eea-055a69f0a1c3@kdbg.org","threadId":"66787","inReplyTo":"20261109-pks-gitlab-ci-speedups-v2-0-508832602d0b@pks.im","subject":"Re: [PATCH v2 0/7] ci: some speedups for slow tests","fromName":"Johannes Sixt","fromEmail":"j6t@kdbg.org","sentAt":"2026-11-12T08:19:22Z","receivedAt":"2026-11-12T08:19:22Z","isPatch":true,"sender":{"key":"j6t@kdbg.org","avatar":"https://avatars.githubusercontent.com/u/14810926?v=4"},"body":"Am 09.11.26 um 10:58 schrieb Patrick Steinhardt:\n> The series contains a couple of bits that are specific to GitLab, and a\n> couple of fixes that are generic. Some highlights:\n> \n>   - Installing the Git for Windows SDK speeds up from 3 minutes to 10\n>     seconds by using tar(1) instead of Extract-Archive.\n> \n>   - t0027 speeds up from >10 minutes to 6 minutes on Windows by avoiding\n>     the overhead of spawning redundant processes.\n> \n>   - t5551 and t5559 both speed up by a factor of 10 on macOS, going from\n>     20 minutes each to 2 minutes each by using reftables in a repo with\n>     100k refs. \"osx-meson\" and \"osx-clang\" thus go from 29 minutes to 18\n>     minutes end-to-end.\n> \n>   - \"fedora-breaking-changes-meson\" speeds up by a factor of 2, going\n>     from 35 minutes to 16 minutes. Turns out our SHA256 implementation\n>     is quite slow with \"-O0\".\n> \n> Unfortunately, our Windows runners at GitLab are still atrociously slow\n> by themselves. So the critical path only improves by about ~10 minutes,\n> going from 60-70 minutes to 53 minutes end-to-end. But hey, at least the\n> other jobs now give feedback a bit faster. And I guess on GitHub it may\n> be a different story because Windows runners are way better there and\n> are typically not on the critical path.\n\n\nI've tested this series on my Windows notebook and it is looking good.\n\nThe runtime of t0027 went down from 4:45 to 3:14, and that of t1092 from\n6:15 to 5:33. t5551 is skipped in my setup.\n\nThanks!\n\n-- Hannes\n\n\n"}]}