threads / patch / 34127
Subject: [PATCH 1/3] rebase: guard against missing files in read_basic_state()
Ramkumar Ramachandra[PATCH 0/3] Fix a couple of edge cases in autostash
Hi,
I apologize for having missed these two trivial cases in the original series.
Ramkumar Ramachandra (3): rebase: guard against missing files in read_basic_state() rebase: finish_rebase() in fast-forward rebase rebase: finish_rebase() in noop rebase
git-rebase.sh | 4 ++++ t/t3420-rebase-autostash.sh | 22 ++++++++++++++++++++++ 2 files changed, 26 insertions(+)
-- 1.8.3.1.381.gf08dd97.dirty
Ramkumar RamachandraSigned-off-by: Ramkumar Ramachandra <artagnon@gmail.com> --- git-rebase.sh | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/git-rebase.sh b/git-rebase.sh index d0c11a9..2122fe0 100755 --- a/git-rebase.sh +++ b/git-rebase.sh @@ -84,6 +84,8 @@ keep_empty= test "$(git config --bool rebase.autosquash)" = "true" && autosquash=t read_basic_state () { + test -f "$state_dir/head-name" && + test -f "$state_dir/onto" && head_name=$(cat "$state_dir"/head-name) && onto=$(cat "$state_dir"/onto) && # We always write to orig-head, but interactive rebase used to write to
-- 1.8.3.1.381.gf08dd97.dirty
Junio C HamanoRe: [PATCH 1/3] rebase: guard against missing files in read_basic_state()
Ramkumar Ramachandra <artagnon@gmail.com> writes:
> Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
This does not affect correctness; i.e. head_name=$(cat that-file) will error out if the file is missing, right?
A more troublesome is that nobody seems to check the return value of this function. If head-name, onto or orig-head is missing, is that an error condition that should make the callers of read_basic_state stop and refuse to proceed?
The way the && cascade is used seems to indicate that, but up to the point where it sents $verbose. If and only if head-name, onto, orig-head and quiet can be read in state-dir, verbose in state-dir is checked and only then $verbose is set.
Martin, this seems to be from your series around early Feburary 2011. Do you recall why these checks are cascaded this way? I do not offhand think of a good reason.
> ---
> git-rebase.sh | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/git-rebase.sh b/git-rebase.sh
> index d0c11a9..2122fe0 100755
> --- a/git-rebase.sh
> +++ b/git-rebase.sh
> @@ -84,6 +84,8 @@ keep_empty=
> test "$(git config --bool rebase.autosquash)" = "true" && autosquash=t
>
> read_basic_state () {
> + test -f "$state_dir/head-name" &&
> + test -f "$state_dir/onto" &&
> head_name=$(cat "$state_dir"/head-name) &&
> onto=$(cat "$state_dir"/onto) &&
> # We always write to orig-head, but interactive rebase used to write to
Martin von ZweigbergkRe: [PATCH 1/3] rebase: guard against missing files in read_basic_state()
On Thu, Jun 13, 2013 at 3:29 PM, Junio C Hamano <gitster@pobox.com> wrote:
> Ramkumar Ramachandra <artagnon@gmail.com> writes: > > A more troublesome is that nobody seems to check the return value of > this function. If head-name, onto or orig-head is missing, is that > an error condition that should make the callers of read_basic_state > stop and refuse to proceed?
Since we unconditionally write those three (and 'quiet'), it seems reasonable to require all of them to be there when continuing, so I think you're right that we should fail fast.
> The way the && cascade is used seems to indicate that, but up to the > point where it sents $verbose. If and only if head-name, onto, orig-head > and quiet can be read in state-dir, verbose in state-dir is checked > and only then $verbose is set. > > Martin, this seems to be from your series around early Feburary > 2011. Do you recall why these checks are cascaded this way? > I do not offhand think of a good reason.
Neither do I. I think the cascading after 'quiet' is just a mistake on my part. The consequences are probably close to none, since if one of earlier commands fail, the other files will probably not be there either. (Not defending it; I'm happy if it gets fixed, e.g. by making it fail fast.)
Junio C HamanoRe: [PATCH 1/3] rebase: guard against missing files in read_basic_state()
Martin von Zweigbergk <martinvonz@gmail.com> writes:
> On Thu, Jun 13, 2013 at 3:29 PM, Junio C Hamano <gitster@pobox.com> wrote: >> Ramkumar Ramachandra <artagnon@gmail.com> writes: >> >> A more troublesome is that nobody seems to check the return value of >> this function. If head-name, onto or orig-head is missing, is that >> an error condition that should make the callers of read_basic_state >> stop and refuse to proceed? > > Since we unconditionally write those three (and 'quiet'), it seems > reasonable to require all of them to be there when continuing, so I > think you're right that we should fail fast. > >> The way the && cascade is used seems to indicate that, but up to the >> point where it sents $verbose. If and only if head-name, onto, orig-head >> and quiet can be read in state-dir, verbose in state-dir is checked >> and only then $verbose is set. >> >> Martin, this seems to be from your series around early Feburary >> 2011. Do you recall why these checks are cascaded this way? >> I do not offhand think of a good reason. > > Neither do I. I think the cascading after 'quiet' is just a mistake on > my part. The consequences are probably close to none, since if one of > earlier commands fail, the other files will probably not be there > either. (Not defending it; I'm happy if it gets fixed, e.g. by making > it fail fast.)
I think this is probably the right thing to do, if we want to honor the original intention of the earlier part of && cascade. Everything before this new "|| die" reads from a file that should always exist (e.g. even when not asked to be quiet, that state is not signaled by the lack of $state_dir/quiet, but by having an empty string in it), while everything after check optional state variable files (e.g. if $state_dir/verbose does not exist, it is not an error, but signals that the user did not ask us to be verbose).
Note that applying this patch _could_ uncover latent bug that was masked by the lack of "die" here (maybe later codepath may depended on not having orig_head at all and the only observable effect was that in such a case, both quiet and verbose were silently ignored, because the control did not reach the GIT_QUIET=... and verbose=t assignments.
git-rebase.sh | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/git-rebase.sh b/git-rebase.sh index d0c11a9..90506ba 100755 --- a/git-rebase.sh +++ b/git-rebase.sh @@ -95,7 +95,9 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + GIT_QUIET=$(cat "$state_dir"/quiet) || + die "failed to read basic rebase state from $state_dir" + test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&
Ramkumar Ramachandra[PATCH 2/3] rebase: finish_rebase() in fast-forward rebase
In the following case
$ git rebase master Fast-forwarded autostash-fix to master.
The autostash is not applied automatically, because this codepath forgets to call finish_rebase(). Fix this. Also add a test to guard against regressions.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> --- git-rebase.sh | 1 + t/t3420-rebase-autostash.sh | 11 +++++++++++ 2 files changed, 12 insertions(+)
git-rebase.sh, t/t3420-rebase-autostash.sh
diff --git a/git-rebase.sh b/git-rebase.sh index 2122fe0..154d4be 100755 --- a/git-rebase.sh +++ b/git-rebase.sh @@ -579,6 +579,7 @@ if test "$mb" = "$orig_head" then say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" move_to_original_branch + finish_rebase exit 0 fi diff --git a/t/t3420-rebase-autostash.sh b/t/t3420-rebase-autostash.sh index 479cbb2..1bde007 100755 --- a/t/t3420-rebase-autostash.sh +++ b/t/t3420-rebase-autostash.sh @@ -141,6 +141,17 @@ testrebase() { ' } +test_expect_success "rebase: fast-forward rebase" ' + test_config rebase.autostash true && + git reset --hard && + git checkout -b behind-feature-branch feature-branch~1 && + test_when_finished git branch -D behind-feature-branch && + echo dirty >>file1 && + git rebase feature-branch && + grep dirty file1 && + git checkout feature-branch +' + testrebase "" .git/rebase-apply testrebase " --merge" .git/rebase-merge testrebase " --interactive" .git/rebase-merge
-- 1.8.3.1.381.gf08dd97.dirty
Ramkumar Ramachandra[PATCH 3/3] rebase: finish_rebase() in noop rebase
In the following case
$ git rebase master Current branch autostash-fix is up to date.
the autostash is not applied automatically, because this codepath forgets to call finish_rebase(). Fix this. Also add a test to guard against regressions.
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com> --- git-rebase.sh | 1 + t/t3420-rebase-autostash.sh | 11 +++++++++++ 2 files changed, 12 insertions(+)
git-rebase.sh, t/t3420-rebase-autostash.sh
diff --git a/git-rebase.sh b/git-rebase.sh index 154d4be..2d5c2bd 100755 --- a/git-rebase.sh +++ b/git-rebase.sh @@ -547,6 +547,7 @@ then # Lazily switch to the target branch if needed... test -z "$switch_to" || git checkout "$switch_to" -- say "$(eval_gettext "Current branch \$branch_name is up to date.")" + finish_rebase exit 0 else say "$(eval_gettext "Current branch \$branch_name is up to date, rebase forced.")" diff --git a/t/t3420-rebase-autostash.sh b/t/t3420-rebase-autostash.sh index 1bde007..90eb264 100755 --- a/t/t3420-rebase-autostash.sh +++ b/t/t3420-rebase-autostash.sh @@ -152,6 +152,17 @@ test_expect_success "rebase: fast-forward rebase" ' git checkout feature-branch ' +test_expect_success "rebase: noop rebase" ' + test_config rebase.autostash true && + git reset --hard && + git checkout -b same-feature-branch feature-branch && + test_when_finished git branch -D same-feature-branch && + echo dirty >>file1 && + git rebase feature-branch && + grep dirty file1 && + git checkout feature-branch +' + testrebase "" .git/rebase-apply testrebase " --merge" .git/rebase-merge testrebase " --interactive" .git/rebase-merge
-- 1.8.3.1.381.gf08dd97.dirty