patchmeson: regenerate config-list.h when Documentation changes

> Le 7 févr. 2026 à 16:59, D. Ben Knoble <ben.knoble+github@gmail.com> a écrit :
> 
> diff --git a/generate-configlist-deps.sh b/generate-configlist-deps.sh
> new file mode 100755
> index 0000000000..a873e976b6
> --- /dev/null
> +++ b/generate-configlist-deps.sh
> @@ -0,0 +1,5 @@
> +#! /bin/sh
> +
> +printf '%s\n' \
> +  "$MESON_SOURCE_ROOT"/Documentation/*config.adoc \
> +  "$MESON_SOURCE_ROOT"/Documentation/config/*.adoc
> diff --git a/meson.build b/meson.build
> index 3a1d12caa4..4df7caa206 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,6 +720,7 @@ endif
> 
> builtin_sources += custom_target(
>   output: 'config-list.h',
> +  depend_files: run_command(meson.current_source_dir() + '/generate-configlist-deps').stdout().strip().split('\n'),
>   

> Notes (benknoble/commits):
>     Changes from v2 (<c9ae171eed6bd5b0fa6671b10a5ad0da024f36d0.1770649805.git.ben.knoble+github@gmail.com>):
>     
>     • Pick up (and tweak) Patrick's depfile proposal
>     • Include the script itself as a dependency
>     • Escape output paths (spaces, octothorpes, and backslashes) for Ninja
>     
>     I'm not 100% sure I've actually done the escaping correctly, though,
>     since Ninja's source says that a space preceded by 2N backslashes
>     represents 2N backslashes at the end of a filename, and

>         λ printf '%s\n' 'foo\' | sed 's/[# \\]/\\&/g' | xxd
>         00000000: 666f 6f5c 5c0a                           foo\\.
>     
>     So would they interpret that as the filename 'foo\\' instead of 'foo\' ?
>     (Or, no because the 2N slashes aren't followed by a SP, but a NL?)

>  generate-configlist.sh | 10 +++++++++-
>  meson.build            |  4 +++-
>  2 files changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..091efd4564 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -36,3 +37,10 @@ print_config_list () {
>  	echo
>  	print_config_list
>  } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +	printf "$OUTPUT: %s\n" "$0" "$SOURCE_DIR"/Documentation/*config.adoc \

> diff --git a/meson.build b/meson.build
> index 3a1d12caa4..fb5d7367f5 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,13 @@ endif
>  
>  builtin_sources += custom_target(
>    output: 'config-list.h',
> +  depfile: 'config-list.h.d',
>    command: [
>      shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>      meson.current_source_dir(),
>      '@OUTPUT@',
> +    '@DEPFILE@',

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list
> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make.

> Helped-by: Patrick Steinhardt <ps@pks.im>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>      Changes from v2 (<c9ae171eed6bd5b0fa6671b10a5ad0da024f36d0.1770649805.git.ben.knoble+github@gmail.com>):
>      
>      • Pick up (and tweak) Patrick's depfile proposal
>      • Include the script itself as a dependency
>      • Escape output paths (spaces, octothorpes, and backslashes) for Ninja
>      
>      I'm not 100% sure I've actually done the escaping correctly, though,
>      since Ninja's source says that a space preceded by 2N backslashes
>      represents 2N backslashes at the end of a filename, and
>      
>          λ printf '%s\n' 'foo\' | sed 's/[# \\]/\\&/g' | xxd
>          00000000: 666f 6f5c 5c0a                           foo\\.
>      
>      So would they interpret that as the filename 'foo\\' instead of 'foo\' ?
>      (Or, no because the 2N slashes aren't followed by a SP, but a NL?)

> +if test -n "$DEPFILE"
> +then
> +	printf "$OUTPUT: %s\n" "$0" "$SOURCE_DIR"/Documentation/*config.adoc \
> +	    "$SOURCE_DIR"/Documentation/config/*.adoc |
> +	    sed 's/[# \\]/\\&/g' >"$DEPFILE"

> +fi
> diff --git a/meson.build b/meson.build
> index 3a1d12caa4..fb5d7367f5 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,13 @@ endif
>   
>   builtin_sources += custom_target(
>     output: 'config-list.h',
> +  depfile: 'config-list.h.d',
>     command: [
>       shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>       meson.current_source_dir(),
>       '@OUTPUT@',
> +    '@DEPFILE@',
>     ],
>     env: script_environment,
>   )
> 
> base-commit: 19a08e0c02faf1c5b4efd1add85598cf8390ff7e
> prerequisite-patch-id: 364ba1899740b93be5957262d3583348d030e8fa

> 
> Le 11 févr. 2026 à 18:51, D. Ben Knoble <ben.knoble+github@gmail.com> a écrit :
> 
> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list
> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make.
> 
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>    Changes from v2 (<c9ae171eed6bd5b0fa6671b10a5ad0da024f36d0.1770649805.git.ben.knoble+github@gmail.com>):
> 
>    • Pick up (and tweak) Patrick's depfile proposal
>    • Include the script itself as a dependency
>    • Escape output paths (spaces, octothorpes, and backslashes) for Ninja
> 
>    I'm not 100% sure I've actually done the escaping correctly, though,
>    since Ninja's source says that a space preceded by 2N backslashes
>    represents 2N backslashes at the end of a filename, and
> 
>        λ printf '%s\n' 'foo\' | sed 's/[# \\]/\\&/g' | xxd
>        00000000: 666f 6f5c 5c0a                           foo\\.
> 
>    So would they interpret that as the filename 'foo\\' instead of 'foo\' ?
>    (Or, no because the 2N slashes aren't followed by a SP, but a NL?)
> 
> generate-configlist.sh | 10 +++++++++-
> meson.build            |  4 +++-
> 2 files changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..091efd4564 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -2,10 +2,11 @@
> 
> SOURCE_DIR="$1"
> OUTPUT="$2"
> +DEPFILE="$3"
> 
> if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
> then
> -    echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
> +    echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT> [<DEPFILE>]"
>    exit 1
> fi
> 
> @@ -36,3 +37,10 @@ print_config_list () {
>    echo
>    print_config_list
> } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +    printf "$OUTPUT: %s\n" "$0" "$SOURCE_DIR"/Documentation/*config.adoc \
> +        "$SOURCE_DIR"/Documentation/config/*.adoc |
> +        sed 's/[# \\]/\\&/g' >"$DEPFILE"

> Le 16 févr. 2026 à 17:29, D. Ben Knoble <ben.knoble+github@gmail.com> a écrit :
> 
> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list
> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make. We assume that if a user adds a new file under
> Documentation/config then they will also edit one of the existing files
> to include that new file, and that will trigger a rebuild.
> 
> Also mark the generator script as a dependency.
> 
> Combining the following commands helps debug dependencies:
> 
>    ninja -C <builddir> -t deps config-list.h
>    ninja -C <builddir> -t browse config-list.h
> 
> The former lists all the dependencies discovered from our output ".d"
> file (the config documentation) and the latter shows the dependency on
> the script itself, among other useful edges in the dependency graph.
> 
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>    Changes from v3 (<0a344f1f3ee4a5d95c6f46df030b9936db4354a1.1770853297.git.ben.knoble+github@gmail.com>):
> 
>    • Include the script itself as a dependency via depfile
>    • Fix output path escaping (spaces, octothorpes; drop backslashes) for
>      Ninja (I've used a loop because I couldn't find a portable construct
>      that could escape only the remainder of the lines in the way I
>      wanted).
>    • Mention our assumptions about Documentation updates triggering
>      rebuilds
>    • Also include some debugging information in the commit message
> 
> generate-configlist.sh | 12 +++++++++++-
> meson.build            |  5 ++++-
> 2 files changed, 15 insertions(+), 2 deletions(-)
> 
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade209..63a2e8b6938 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -2,10 +2,11 @@
> 
> SOURCE_DIR="$1"
> OUTPUT="$2"
> +DEPFILE="$3"
> 
> if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
> then
> -    echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
> +    echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT> [<DEPFILE>]"
>    exit 1
> fi
> 
> @@ -36,3 +37,12 @@ print_config_list ()
>    echo
>    print_config_list
> } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +    for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
> +        "$SOURCE_DIR"/Documentation/config/*.adoc
> +    do
> +        printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"
> +    done >"$DEPFILE"
> +fi
> diff --git a/meson.build b/meson.build
> index 3a1d12caa4b..e4b8f1e33d2 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,14 @@ endif
> 
> builtin_sources += custom_target(
>   output: 'config-list.h',
> +  depfile: 'config-list.h.d',
> +  depend_files: [ 'generate-configlist.sh' ],
>   command: [
>     shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>     meson.current_source_dir(),
>     '@OUTPUT@',
> +    '@DEPFILE@',
>   ],
>   env: script_environment,
> )
> 
> Diff-intervalle contre v3 :
> 1:  0a344f1f3ee ! 1:  9cdcc9de04f meson: regenerate config-list.h when Documentation changes
>    @@ Commit message
>         Teach the meson build to depend on the Documentation files that
>         generate-configlist.sh reads by having it an additional output as a list
>         of dependency files, since Meson does not have (or want) builtin support
>    -    for globbing like Make.
>    +    for globbing like Make. We assume that if a user adds a new file under
>    +    Documentation/config then they will also edit one of the existing files
>    +    to include that new file, and that will trigger a rebuild.
>    +
>    +    Also mark the generator script as a dependency.
>    +
>    +    Combining the following commands helps debug dependencies:
>    +
>    +        ninja -C <builddir> -t deps config-list.h
>    +        ninja -C <builddir> -t browse config-list.h
>    +
>    +    The former lists all the dependencies discovered from our output ".d"
>    +    file (the config documentation) and the latter shows the dependency on
>    +    the script itself, among other useful edges in the dependency graph.
> 
>         Helped-by: Patrick Steinhardt <ps@pks.im>
>    +    Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> 
> 
>      ## Notes (benknoble/commits) ##
>    -    Changes from v2 (<c9ae171eed6bd5b0fa6671b10a5ad0da024f36d0.1770649805.git.ben.knoble+github@gmail.com>):
>    +    Changes from v3 (<0a344f1f3ee4a5d95c6f46df030b9936db4354a1.1770853297.git.ben.knoble+github@gmail.com>):
> 
>    -    • Pick up (and tweak) Patrick's depfile proposal
>    -    • Include the script itself as a dependency
>    -    • Escape output paths (spaces, octothorpes, and backslashes) for Ninja
>    -
>    -    I'm not 100% sure I've actually done the escaping correctly, though,
>    -    since Ninja's source says that a space preceded by 2N backslashes
>    -    represents 2N backslashes at the end of a filename, and
>    -
>    -        λ printf '%s\n' 'foo\' | sed 's/[# \\]/\\&/g' | xxd
>    -        00000000: 666f 6f5c 5c0a                           foo\\.
>    -
>    -    So would they interpret that as the filename 'foo\\' instead of 'foo\' ?
>    -    (Or, no because the 2N slashes aren't followed by a SP, but a NL?)
>    +    • Include the script itself as a dependency via depfile
>    +    • Fix output path escaping (spaces, octothorpes; drop backslashes) for
>    +      Ninja (I've used a loop because I couldn't find a portable construct
>    +      that could escape only the remainder of the lines in the way I
>    +      wanted).
>    +    • Mention our assumptions about Documentation updates triggering
>    +      rebuilds
>    +    • Also include some debugging information in the commit message
> 
>      ## generate-configlist.sh ##
>     @@
>    @@ generate-configlist.sh: print_config_list ()
>     +
>     +if test -n "$DEPFILE"
>     +then
>    -+    printf "$OUTPUT: %s\n" "$0" "$SOURCE_DIR"/Documentation/*config.adoc \
>    -+        "$SOURCE_DIR"/Documentation/config/*.adoc |
>    -+        sed 's/[# \\]/\\&/g' >"$DEPFILE"
>    ++    for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
>    ++        "$SOURCE_DIR"/Documentation/config/*.adoc
>    ++    do
>    ++        printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"
>    ++    done >"$DEPFILE"
>     +fi
> 
>      ## meson.build ##
>    @@ meson.build: endif
>      builtin_sources += custom_target(
>        output: 'config-list.h',
>     +  depfile: 'config-list.h.d',
>    ++  depend_files: [ 'generate-configlist.sh' ],
>        command: [
>          shell,
>     -    meson.current_source_dir() + '/generate-configlist.sh',
> 
> base-commit: 19a08e0c02faf1c5b4efd1add85598cf8390ff7e
> prerequisite-patch-id: 364ba1899740b93be5957262d3583348d030e8fa
> --
> 2.48.1

>
> On Mon, Feb 16, 2026 at 07:33:19PM -0500, Ben Knoble wrote:
> > Hmm. I’m pretty sure this worked locally but I’m seeing some CI
> > failures on GitHub. Will investigate.
>
> Where did you see the failures? The changes look good to me, and things
> work as expected on my machine, so I'm wondering what the root cause is.
> A test run at [1] in GitLab's CI is also successful (ignore the
> MSVC+Meson failures, a fix for these is being reviewed on the mailing
> list already).
>
> Patrick
>
> [1]: https://gitlab.com/gitlab-org/git/-/merge_requests/505

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list
> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make. We assume that if a user adds a new file under
> Documentation/config then they will also edit one of the existing files
> to include that new file, and that will trigger a rebuild.
> 
> Also mark the generator script as a dependency.
> 
> Combining the following commands helps debug dependencies:
> 
>     ninja -C <builddir> -t deps config-list.h
>     ninja -C <builddir> -t browse config-list.h
> 
> The former lists all the dependencies discovered from our output ".d"
> file (the config documentation) and the latter shows the dependency on
> the script itself, among other useful edges in the dependency graph.

> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade209..63a2e8b6938 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -36,3 +37,12 @@ print_config_list ()
>  	echo
>  	print_config_list
>  } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +	for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc
> +	do
> +		printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"

> diff --git a/meson.build b/meson.build
> index 3a1d12caa4b..e4b8f1e33d2 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,14 @@ endif
>  
>  builtin_sources += custom_target(
>    output: 'config-list.h',
> +  depfile: 'config-list.h.d',
> +  depend_files: [ 'generate-configlist.sh' ],
>    command: [
>      shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>      meson.current_source_dir(),
>      '@OUTPUT@',
> +    '@DEPFILE@',
>    ],
>    env: script_environment,
>  )

>
> On Mon, Feb 16, 2026 at 05:28:57PM -0500, D. Ben Knoble wrote:
> > The Meson-based build doesn't know when to rebuild config-list.h, so the
> > header is sometimes stale.
> >
> > For example, an old build directory might have config-list.h from before
> > 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> > 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> > it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> > from sources fixes the artifact and the test.
> >
> > Teach the meson build to depend on the Documentation files that
> > generate-configlist.sh reads by having it an additional output as a list
> > of dependency files, since Meson does not have (or want) builtin support
> > for globbing like Make. We assume that if a user adds a new file under
> > Documentation/config then they will also edit one of the existing files
> > to include that new file, and that will trigger a rebuild.
> >
> > Also mark the generator script as a dependency.
> >
> > Combining the following commands helps debug dependencies:
> >
> >     ninja -C <builddir> -t deps config-list.h
> >     ninja -C <builddir> -t browse config-list.h
> >
> > The former lists all the dependencies discovered from our output ".d"
> > file (the config documentation) and the latter shows the dependency on
> > the script itself, among other useful edges in the dependency graph.
>
> Good hint, thanks for putting it here.
>
> > diff --git a/generate-configlist.sh b/generate-configlist.sh
> > index 75c39ade209..63a2e8b6938 100755
> > --- a/generate-configlist.sh
> > +++ b/generate-configlist.sh
> > @@ -36,3 +37,12 @@ print_config_list ()
> >       echo
> >       print_config_list
> >  } >"$OUTPUT"
> > +
> > +if test -n "$DEPFILE"
> > +then
> > +     for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
> > +             "$SOURCE_DIR"/Documentation/config/*.adoc
> > +     do
> > +             printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"
>
> Tiny nit: can't we simplify this to "$(echo "$doc" | sed ...)"?

> > diff --git a/meson.build b/meson.build
> > index 3a1d12caa4b..e4b8f1e33d2 100644
> > --- a/meson.build
> > +++ b/meson.build
> > @@ -720,11 +720,14 @@ endif
> >
> >  builtin_sources += custom_target(
> >    output: 'config-list.h',
> > +  depfile: 'config-list.h.d',
> > +  depend_files: [ 'generate-configlist.sh' ],
> >    command: [
> >      shell,
> > -    meson.current_source_dir() + '/generate-configlist.sh',
> > +    meson.current_source_dir() / 'generate-configlist.sh',
> >      meson.current_source_dir(),
> >      '@OUTPUT@',
> > +    '@DEPFILE@',
> >    ],
> >    env: script_environment,
> >  )
>
> And this looks as expected now, too. I also verified locally that this
> correctly rebuilds things when either the script or any of the files
> change.
>
> I wonder whether we also want to have a second patch on top to adjust
> our Makefile? E.g. something like this:
>
> diff --git a/Makefile b/Makefile
> index 47ed9fa7fd..ed82775ec2 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -2688,8 +2688,10 @@ $(BUILT_INS): git$X
>
>  config-list.h: generate-configlist.sh
>
> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
> -       $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
> +config-list.h: generate-configlist.sh
> +       $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
> +
> +-include .depend/config-list.h.d
>
>  command-list.h: generate-cmdlist.sh command-list.txt
>
>
> Please feel free to skip this part though.
>
> Thanks!
>
> Patrick

>> > +     for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
>> > +             "$SOURCE_DIR"/Documentation/config/*.adoc
>> > +     do
>> > +             printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"
>>
>> Tiny nit: can't we simplify this to "$(echo "$doc" | sed ...)"?
>
> Given how unportable echo is [1], I'd prefer to keep printf.
>
> [1]: https://github.com/benknoble/echocho, for one. If the doc started
> with "-e" for example the results might be unreliable.

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list
> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make. We assume that if a user adds a new file under
> Documentation/config then they will also edit one of the existing files
> to include that new file, and that will trigger a rebuild.
> 
> Also mark the generator script as a dependency.
> 
> Combining the following commands helps debug dependencies:
> 
>      ninja -C <builddir> -t deps config-list.h
>      ninja -C <builddir> -t browse config-list.h
> 
> The former lists all the dependencies discovered from our output ".d"
> file (the config documentation) and the latter shows the dependency on
> the script itself, among other useful edges in the dependency graph.
> 
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>

> +if test -n "$DEPFILE"
> +then
> +	for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc
> +	do
> +		printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"\\

> On 16/02/2026 22:28, D. Ben Knoble wrote:
> > +if test -n "$DEPFILE"
> > +then
> > +     for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
> > +             "$SOURCE_DIR"/Documentation/config/*.adoc
> > +     do
> > +             printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"\\
>
> This forks two processes for every file which is a bit inefficient and
> will be especially slow on windows.

> If we quote $OUTPUT first we should
> be able to use a single sed process for all the files
>
>         QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT"| sed 's|[/\]|\\&|g')"
>         printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
>                 "$SOURCE_DIR"/Documentation/config/*.adoc |
>         sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"

> On Tue, Feb 17, 2026 at 4:20 AM Phillip Wood <phillip.wood123@gmail.com> wrote:
>> On 16/02/2026 22:28, D. Ben Knoble wrote:
>>> +if test -n "$DEPFILE"
>>> +then
>>> +     for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
>>> +             "$SOURCE_DIR"/Documentation/config/*.adoc
>>> +     do
>>> +             printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"\\
>>
>> This forks two processes for every file which is a bit inefficient and
>> will be especially slow on windows.
> 
> Yeah. I couldn't find a way I trusted to behave in the presence of
> strange characters, though…
> 
>> If we quote $OUTPUT first we should
>> be able to use a single sed process for all the files
>>
>>          QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT"| sed 's|[/\]|\\&|g')"
>>          printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
>>                  "$SOURCE_DIR"/Documentation/config/*.adoc |
>>          sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> 
> …and this version matches a concept I considered. Let me think aloud.
> 
> To quote output, we backslash-escape any "/" and "\"; we are going to
> use it in the replacement side of sed's substitute command with "/"
> delimiters.
> 
> My manual says "&" and "\[0-9]" are also special in the replacement
> string.

> complains about bad backreferences in cases like
> 
>      echo abc | sed 's/^/foo\1/'
> 
> OTOH, escaping backslashes already covers such backreferences. So I
> think we'd need something closer to
> 
>      sed 's,[&/\],\\&,g'
> 
> ? If you agree with that reasoning, I'll include this in the next
> version. (I'll also try this out a bit first.)

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list
> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make. We assume that if a user adds a new file under
> Documentation/config then they will also edit one of the existing files
> to include that new file, and that will trigger a rebuild. Also mark the
> generator script as a dependency.
> 
> While we're at it, teach the Makefile to use the same "the script knows
> it's dependencies" logic.
> 
> For Meson, combining the following commands helps debug dependencies:
> 
>      ninja -C <builddir> -t deps config-list.h
>      ninja -C <builddir> -t browse config-list.h
> 
> The former lists all the dependencies discovered from our output ".d"
> file (the config documentation) and the latter shows the dependency on
> the script itself, among other useful edges in the dependency graph.
> 
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>      Changes from v4 (<9cdcc9de04f0f8fff657f0474b31c063466ed808.1771280837.git.ben.knoble+github@gmail.com>):
>      
>      • Include Patrick's suggested Makefile changes. Note there's no quiet
>        equivalent for mdkir that isn't for the current target's containing
>        directory…
>      • Make depfile output efficient again, thanks to Phillip.
>      
>      I've kept printf instead of echo (from Patrick/Junio) because I think it
>      is easier to reason about ("it works" vs. "did I use this in a way that
>      might cause problems").
>      
>      Junio asked about other problematic bytes: the other one I could think
>      of (since all the inputs should be paths, anyway) is newlines. I gave
>      meson's depfile.py a glance [1], and it looks like they don't handle
>      newlines in paths. Other whitespace doesn't appear to be an issue (see
>      "elif c in {' ', '\n'}"); I think _most_ characters are just added to
>      the filename.
>      
>      [1]: https://github.com/mesonbuild/meson/blob/master/mesonbuild/depfile.py
> 
>   Makefile               |  5 +++--
>   generate-configlist.sh | 11 ++++++++++-
>   meson.build            |  5 ++++-
>   3 files changed, 17 insertions(+), 4 deletions(-)
> 
> diff --git a/Makefile b/Makefile
> index 7f37ad8f58..6f926ffb1f 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
>   	cp $< $@
>   
>   config-list.h: generate-configlist.sh
> +	@mkdir -p .depend
> +	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>   
> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
> -	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
> +-include .depend/config-list.h.d
>   
>   command-list.h: generate-cmdlist.sh command-list.txt
>   
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..39ac8845ab 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -2,10 +2,11 @@
>   
>   SOURCE_DIR="$1"
>   OUTPUT="$2"
> +DEPFILE="$3"
>   
>   if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
>   then
> -	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
> +	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT> [<DEPFILE>]"
>   	exit 1
>   fi
>   
> @@ -36,3 +37,11 @@ print_config_list () {
>   	echo
>   	print_config_list
>   } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> +	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc |
> +		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> +fi
> diff --git a/meson.build b/meson.build
> index 762e2d0fc0..74b459b004 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,14 @@ endif
>   
>   builtin_sources += custom_target(
>     output: 'config-list.h',
> +  depfile: 'config-list.h.d',
> +  depend_files: [ 'generate-configlist.sh' ],
>     command: [
>       shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>       meson.current_source_dir(),
>       '@OUTPUT@',
> +    '@DEPFILE@',
>     ],
>     env: script_environment,
>   )
> 
> Diff-intervalle contre v4 :
> 1:  e2f4e1f9ba < -:  ---------- completion: add stash import, export
> 2:  9cdcc9de04 ! 1:  611a94cd98 meson: regenerate config-list.h when Documentation changes
>      @@ Metadata
>       Author: D. Ben Knoble <ben.knoble+github@gmail.com>
>       
>        ## Commit message ##
>      -    meson: regenerate config-list.h when Documentation changes
>      +    build: regenerate config-list.h when Documentation changes
>       
>           The Meson-based build doesn't know when to rebuild config-list.h, so the
>           header is sometimes stale.
>      @@ Commit message
>           of dependency files, since Meson does not have (or want) builtin support
>           for globbing like Make. We assume that if a user adds a new file under
>           Documentation/config then they will also edit one of the existing files
>      -    to include that new file, and that will trigger a rebuild.
>      +    to include that new file, and that will trigger a rebuild. Also mark the
>      +    generator script as a dependency.
>       
>      -    Also mark the generator script as a dependency.
>      +    While we're at it, teach the Makefile to use the same "the script knows
>      +    it's dependencies" logic.
>       
>      -    Combining the following commands helps debug dependencies:
>      +    For Meson, combining the following commands helps debug dependencies:
>       
>               ninja -C <builddir> -t deps config-list.h
>               ninja -C <builddir> -t browse config-list.h
>      @@ Commit message
>       
>       
>        ## Notes (benknoble/commits) ##
>      -    Changes from v3 (<0a344f1f3ee4a5d95c6f46df030b9936db4354a1.1770853297.git.ben.knoble+github@gmail.com>):
>      +    Changes from v4 (<9cdcc9de04f0f8fff657f0474b31c063466ed808.1771280837.git.ben.knoble+github@gmail.com>):
>       
>      -    • Include the script itself as a dependency via depfile
>      -    • Fix output path escaping (spaces, octothorpes; drop backslashes) for
>      -      Ninja (I've used a loop because I couldn't find a portable construct
>      -      that could escape only the remainder of the lines in the way I
>      -      wanted).
>      -    • Mention our assumptions about Documentation updates triggering
>      -      rebuilds
>      -    • Also include some debugging information in the commit message
>      +    • Include Patrick's suggested Makefile changes. Note there's no quiet
>      +      equivalent for mdkir that isn't for the current target's containing
>      +      directory…
>      +    • Make depfile output efficient again, thanks to Phillip.
>      +
>      +    I've kept printf instead of echo (from Patrick/Junio) because I think it
>      +    is easier to reason about ("it works" vs. "did I use this in a way that
>      +    might cause problems").
>      +
>      +    Junio asked about other problematic bytes: the other one I could think
>      +    of (since all the inputs should be paths, anyway) is newlines. I gave
>      +    meson's depfile.py a glance [1], and it looks like they don't handle
>      +    newlines in paths. Other whitespace doesn't appear to be an issue (see
>      +    "elif c in {' ', '\n'}"); I think _most_ characters are just added to
>      +    the filename.
>      +
>      +    [1]: https://github.com/mesonbuild/meson/blob/master/mesonbuild/depfile.py
>      +
>      + ## Makefile ##
>      +@@ Makefile: $(BUILT_INS): git$X
>      + 	cp $< $@
>      +
>      + config-list.h: generate-configlist.sh
>      ++	@mkdir -p .depend
>      ++	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>      +
>      +-config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
>      +-	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
>      ++-include .depend/config-list.h.d
>      +
>      + command-list.h: generate-cmdlist.sh command-list.txt
>      +
>       
>        ## generate-configlist.sh ##
>       @@
>      @@ generate-configlist.sh: print_config_list () {
>       +
>       +if test -n "$DEPFILE"
>       +then
>      -+	for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
>      -+		"$SOURCE_DIR"/Documentation/config/*.adoc
>      -+	do
>      -+		printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"
>      -+	done >"$DEPFILE"
>      ++	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
>      ++	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
>      ++		"$SOURCE_DIR"/Documentation/config/*.adoc |
>      ++		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
>       +fi
>       
>        ## meson.build ##
> 
> base-commit: f7e9f6c205466443107228e036b20acb7baa8c50

>
> Hi Ben
>
> I tested the meson changes by setting up source and build directories
> with daft names to test the quoting.
>
> $ git worktree add --detach '/dev/shm/s#r#c dir'
> $ cd '/dev/shm/s#r#c dir'
> $ meson setup '/dev/shm/b#u&i ld'
>
> If I build git and then remove Documentation/config/add.adoc
> config-list.h is regenerated, it is not regenerated if I do not change
> any of the config documentation files so it looks to be working
> correctly. I've not tested the Makefile changes.
>
> Thanks for working on it
>
> Phillip

> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..39ac8845ab 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -36,3 +37,11 @@ print_config_list () {
>  	echo
>  	print_config_list
>  } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"

> +then
> +	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> +	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc |
> +		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> +fi

> diff --git a/meson.build b/meson.build
> index 762e2d0fc0..74b459b004 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,14 @@ endif
>  
>  builtin_sources += custom_target(
>    output: 'config-list.h',
> +  depfile: 'config-list.h.d',
> +  depend_files: [ 'generate-configlist.sh' ],
>    command: [
>      shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>      meson.current_source_dir(),
>      '@OUTPUT@',
> +    '@DEPFILE@',
>    ],
>    env: script_environment,
>  )
> 
> Diff-intervalle contre v4 :
> 1:  e2f4e1f9ba < -:  ---------- completion: add stash import, export

>
> On Wed, Feb 18, 2026 at 09:37:59AM -0500, D. Ben Knoble wrote:
> > diff --git a/generate-configlist.sh b/generate-configlist.sh
> > index 75c39ade20..39ac8845ab 100755
> > --- a/generate-configlist.sh
> > +++ b/generate-configlist.sh
> > @@ -36,3 +37,11 @@ print_config_list () {
> >       echo
> >       print_config_list
> >  } >"$OUTPUT"
> > +
> > +if test -n "$DEPFILE"
>
> We could arguably drop conditional now that both the Makefile and Meson
> pass that argument, but I don't mind it much.

> > +then
> > +     QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> > +     printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> > +             "$SOURCE_DIR"/Documentation/config/*.adoc |
> > +             sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> > +fi
>
> Okay, so we first quote the target output file once, and then we quote
> the sources with a single call to sed(1), too. To not perform quoting
> of the target output file twice we don't include it in the printf call,
> but instead prepend it to each line.
>
> Looks reasonable to me, and we spawn at most two sed(1) processes. Nice.
>
> > diff --git a/meson.build b/meson.build
> > index 762e2d0fc0..74b459b004 100644
> > --- a/meson.build
> > +++ b/meson.build
> > @@ -720,11 +720,14 @@ endif
> >
> >  builtin_sources += custom_target(
> >    output: 'config-list.h',
> > +  depfile: 'config-list.h.d',
> > +  depend_files: [ 'generate-configlist.sh' ],
> >    command: [
> >      shell,
> > -    meson.current_source_dir() + '/generate-configlist.sh',
> > +    meson.current_source_dir() / 'generate-configlist.sh',
> >      meson.current_source_dir(),
> >      '@OUTPUT@',
> > +    '@DEPFILE@',
> >    ],
> >    env: script_environment,
> >  )
> >
> > Diff-intervalle contre v4 :
> > 1:  e2f4e1f9ba < -:  ---------- completion: add stash import, export
>
> The range-diff is a bit broken, but that doesn't really matter much.

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it an additional output as a list> of dependency files, since Meson does not have (or want) builtin support
> for globbing like Make.

> We assume that if a user adds a new file under
> Documentation/config then they will also edit one of the existing files
> to include that new file, and that will trigger a rebuild. Also mark the
> generator script as a dependency.
> 
> While we're at it, teach the Makefile to use the same "the script knows
> it's dependencies" logic.
> 
> For Meson, combining the following commands helps debug dependencies:
> 
>      ninja -C <builddir> -t deps config-list.h
>      ninja -C <builddir> -t browse config-list.h
> 
> The former lists all the dependencies discovered from our output ".d"
> file (the config documentation) and the latter shows the dependency on
> the script itself, among other useful edges in the dependency graph.
> 
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>      Changes from v4 (<9cdcc9de04f0f8fff657f0474b31c063466ed808.1771280837.git.ben.knoble+github@gmail.com>):
>      
>      • Include Patrick's suggested Makefile changes. Note there's no quiet
>        equivalent for mdkir that isn't for the current target's containing
>        directory…
>      • Make depfile output efficient again, thanks to Phillip.
>      
>      I've kept printf instead of echo (from Patrick/Junio) because I think it
>      is easier to reason about ("it works" vs. "did I use this in a way that
>      might cause problems").
>      
>      Junio asked about other problematic bytes: the other one I could think
>      of (since all the inputs should be paths, anyway) is newlines. I gave
>      meson's depfile.py a glance [1], and it looks like they don't handle
>      newlines in paths. Other whitespace doesn't appear to be an issue (see
>      "elif c in {' ', '\n'}"); I think _most_ characters are just added to
>      the filename.
>      
>      [1]: https://github.com/mesonbuild/meson/blob/master/mesonbuild/depfile.py
> 
>   Makefile               |  5 +++--
>   generate-configlist.sh | 11 ++++++++++-
>   meson.build            |  5 ++++-
>   3 files changed, 17 insertions(+), 4 deletions(-)
> 
> diff --git a/Makefile b/Makefile
> index 7f37ad8f58..6f926ffb1f 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
>   	cp $< $@
>   
>   config-list.h: generate-configlist.sh
> +	@mkdir -p .depend
> +	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>   
> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
> -	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
> +-include .depend/config-list.h.d
>   
>   command-list.h: generate-cmdlist.sh command-list.txt
>   
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..39ac8845ab 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -2,10 +2,11 @@
>   
>   SOURCE_DIR="$1"
>   OUTPUT="$2"
> +DEPFILE="$3"
>   
>   if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
>   then
> -	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
> +	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT> [<DEPFILE>]"
>   	exit 1
>   fi
>   
> @@ -36,3 +37,11 @@ print_config_list () {
>   	echo
>   	print_config_list
>   } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> +	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc |
> +		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> +fi
> diff --git a/meson.build b/meson.build
> index 762e2d0fc0..74b459b004 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,14 @@ endif
>   
>   builtin_sources += custom_target(
>     output: 'config-list.h',
> +  depfile: 'config-list.h.d',
> +  depend_files: [ 'generate-configlist.sh' ],
>     command: [
>       shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>       meson.current_source_dir(),
>       '@OUTPUT@',
> +    '@DEPFILE@',
>     ],
>     env: script_environment,
>   )
> 
> Diff-intervalle contre v4 :
> 1:  e2f4e1f9ba < -:  ---------- completion: add stash import, export
> 2:  9cdcc9de04 ! 1:  611a94cd98 meson: regenerate config-list.h when Documentation changes
>      @@ Metadata
>       Author: D. Ben Knoble <ben.knoble+github@gmail.com>
>       
>        ## Commit message ##
>      -    meson: regenerate config-list.h when Documentation changes
>      +    build: regenerate config-list.h when Documentation changes
>       
>           The Meson-based build doesn't know when to rebuild config-list.h, so the
>           header is sometimes stale.
>      @@ Commit message
>           of dependency files, since Meson does not have (or want) builtin support
>           for globbing like Make. We assume that if a user adds a new file under
>           Documentation/config then they will also edit one of the existing files
>      -    to include that new file, and that will trigger a rebuild.
>      +    to include that new file, and that will trigger a rebuild. Also mark the
>      +    generator script as a dependency.
>       
>      -    Also mark the generator script as a dependency.
>      +    While we're at it, teach the Makefile to use the same "the script knows
>      +    it's dependencies" logic.
>       
>      -    Combining the following commands helps debug dependencies:
>      +    For Meson, combining the following commands helps debug dependencies:
>       
>               ninja -C <builddir> -t deps config-list.h
>               ninja -C <builddir> -t browse config-list.h
>      @@ Commit message
>       
>       
>        ## Notes (benknoble/commits) ##
>      -    Changes from v3 (<0a344f1f3ee4a5d95c6f46df030b9936db4354a1.1770853297.git.ben.knoble+github@gmail.com>):
>      +    Changes from v4 (<9cdcc9de04f0f8fff657f0474b31c063466ed808.1771280837.git.ben.knoble+github@gmail.com>):
>       
>      -    • Include the script itself as a dependency via depfile
>      -    • Fix output path escaping (spaces, octothorpes; drop backslashes) for
>      -      Ninja (I've used a loop because I couldn't find a portable construct
>      -      that could escape only the remainder of the lines in the way I
>      -      wanted).
>      -    • Mention our assumptions about Documentation updates triggering
>      -      rebuilds
>      -    • Also include some debugging information in the commit message
>      +    • Include Patrick's suggested Makefile changes. Note there's no quiet
>      +      equivalent for mdkir that isn't for the current target's containing
>      +      directory…
>      +    • Make depfile output efficient again, thanks to Phillip.
>      +
>      +    I've kept printf instead of echo (from Patrick/Junio) because I think it
>      +    is easier to reason about ("it works" vs. "did I use this in a way that
>      +    might cause problems").
>      +
>      +    Junio asked about other problematic bytes: the other one I could think
>      +    of (since all the inputs should be paths, anyway) is newlines. I gave
>      +    meson's depfile.py a glance [1], and it looks like they don't handle
>      +    newlines in paths. Other whitespace doesn't appear to be an issue (see
>      +    "elif c in {' ', '\n'}"); I think _most_ characters are just added to
>      +    the filename.
>      +
>      +    [1]: https://github.com/mesonbuild/meson/blob/master/mesonbuild/depfile.py
>      +
>      + ## Makefile ##
>      +@@ Makefile: $(BUILT_INS): git$X
>      + 	cp $< $@
>      +
>      + config-list.h: generate-configlist.sh
>      ++	@mkdir -p .depend
>      ++	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>      +
>      +-config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
>      +-	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
>      ++-include .depend/config-list.h.d
>      +
>      + command-list.h: generate-cmdlist.sh command-list.txt
>      +
>       
>        ## generate-configlist.sh ##
>       @@
>      @@ generate-configlist.sh: print_config_list () {
>       +
>       +if test -n "$DEPFILE"
>       +then
>      -+	for doc in "$SOURCE_DIR"/Documentation/*config.adoc \
>      -+		"$SOURCE_DIR"/Documentation/config/*.adoc
>      -+	do
>      -+		printf "$OUTPUT: %s\n" "$(printf '%s\n' "$doc" | sed 's/[# ]/\\&/g')"
>      -+	done >"$DEPFILE"
>      ++	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
>      ++	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
>      ++		"$SOURCE_DIR"/Documentation/config/*.adoc |
>      ++		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
>       +fi
>       
>        ## meson.build ##
> 
> base-commit: f7e9f6c205466443107228e036b20acb7baa8c50

>
>
> On 2026-02-18 07:37, D. Ben Knoble wrote:
> > The Meson-based build doesn't know when to rebuild config-list.h, so the
> > header is sometimes stale.
> >
> > For example, an old build directory might have config-list.h from before
> > 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> > 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> > it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> > from sources fixes the artifact and the test.
> >
> > Teach the meson build to depend on the Documentation files that
> > generate-configlist.sh reads by having it an additional output as a list
> > of dependency files, since Meson does not have (or want) builtin support
> > for globbing like Make.
>
> That sentence doesn't parse.  Maybe this?
>
> Since Meson does not have (or want) builtin support for globbing like
> Make, teach generate-configlist.sh to generate a dependency list of the
> Documentation files it reads, and have the Meson build incorporate that.
>
>                 M.

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Since Meson does not have (or want) builtin support for globbing like
> Make, teach generate-configlist.sh to also generate a list of
> Documentation files its output depends on, and incorporate that into the
> Meson build.
> 
> We assume that if a user adds a new file under
> Documentation/config then they will also edit one of the existing files
> to include that new file, and that will trigger a rebuild. Also mark the
> generator script as a dependency.
> 
> While we're at it, teach the Makefile to use the same "the script knows
> it's dependencies" logic.
> 
> For Meson, combining the following commands helps debug dependencies:
> 
>     ninja -C <builddir> -t deps config-list.h
>     ninja -C <builddir> -t browse config-list.h
> 
> The former lists all the dependencies discovered from our output ".d"
> file (the config documentation) and the latter shows the dependency on
> the script itself, among other useful edges in the dependency graph.
> 
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>     Changes from v5 (<611a94cd988e3795bc63dba2f1b270aa0d058bd2.1771425395.git.ben.knoble+github@gmail.com>):
>     
>     • Reword a confusing sentence in the commit message
> 
>  Makefile               |  5 +++--
>  generate-configlist.sh | 11 ++++++++++-
>  meson.build            |  5 ++++-
>  3 files changed, 17 insertions(+), 4 deletions(-)
> 
> diff --git a/Makefile b/Makefile
> index 7f37ad8f58..6f926ffb1f 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
>  	cp $< $@
>  
>  config-list.h: generate-configlist.sh
> +	@mkdir -p .depend
> +	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>  
> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
> -	$(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
> +-include .depend/config-list.h.d

>  command-list.h: generate-cmdlist.sh command-list.txt
>  
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..39ac8845ab 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -2,10 +2,11 @@
>  
>  SOURCE_DIR="$1"
>  OUTPUT="$2"
> +DEPFILE="$3"
>  
>  if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
>  then
> -	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
> +	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT> [<DEPFILE>]"
>  	exit 1
>  fi
>  
> @@ -36,3 +37,11 @@ print_config_list () {
>  	echo
>  	print_config_list
>  } >"$OUTPUT"
> +
> +if test -n "$DEPFILE"
> +then
> +	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> +	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc |
> +		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> +fi
> diff --git a/meson.build b/meson.build
> index 762e2d0fc0..74b459b004 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -720,11 +720,14 @@ endif
>  
>  builtin_sources += custom_target(
>    output: 'config-list.h',
> +  depfile: 'config-list.h.d',
> +  depend_files: [ 'generate-configlist.sh' ],
>    command: [
>      shell,
> -    meson.current_source_dir() + '/generate-configlist.sh',
> +    meson.current_source_dir() / 'generate-configlist.sh',
>      meson.current_source_dir(),
>      '@OUTPUT@',
> +    '@DEPFILE@',
>    ],
>    env: script_environment,
>  )
> 
> Diff-intervalle contre v5 :
> 1:  611a94cd98 ! 1:  5dcd4e9308 build: regenerate config-list.h when Documentation changes
>     @@ Commit message
>          it, t9902-completion.sh fails. Regenerating the config-list.h artifact
>          from sources fixes the artifact and the test.
>      
>     -    Teach the meson build to depend on the Documentation files that
>     -    generate-configlist.sh reads by having it an additional output as a list
>     -    of dependency files, since Meson does not have (or want) builtin support
>     -    for globbing like Make. We assume that if a user adds a new file under
>     +    Since Meson does not have (or want) builtin support for globbing like
>     +    Make, teach generate-configlist.sh to also generate a list of
>     +    Documentation files its output depends on, and incorporate that into the
>     +    Meson build.
>     +
>     +    We assume that if a user adds a new file under
>          Documentation/config then they will also edit one of the existing files
>          to include that new file, and that will trigger a rebuild. Also mark the
>          generator script as a dependency.
>     @@ Commit message
>      
>      
>       ## Notes (benknoble/commits) ##
>     -    Changes from v4 (<9cdcc9de04f0f8fff657f0474b31c063466ed808.1771280837.git.ben.knoble+github@gmail.com>):
>     +    Changes from v5 (<611a94cd988e3795bc63dba2f1b270aa0d058bd2.1771425395.git.ben.knoble+github@gmail.com>):
>      
>     -    • Include Patrick's suggested Makefile changes. Note there's no quiet
>     -      equivalent for mdkir that isn't for the current target's containing
>     -      directory…
>     -    • Make depfile output efficient again, thanks to Phillip.
>     -
>     -    I've kept printf instead of echo (from Patrick/Junio) because I think it
>     -    is easier to reason about ("it works" vs. "did I use this in a way that
>     -    might cause problems").
>     -
>     -    Junio asked about other problematic bytes: the other one I could think
>     -    of (since all the inputs should be paths, anyway) is newlines. I gave
>     -    meson's depfile.py a glance [1], and it looks like they don't handle
>     -    newlines in paths. Other whitespace doesn't appear to be an issue (see
>     -    "elif c in {' ', '\n'}"); I think _most_ characters are just added to
>     -    the filename.
>     -
>     -    [1]: https://github.com/mesonbuild/meson/blob/master/mesonbuild/depfile.py
>     +    • Reword a confusing sentence in the commit message
>      
>       ## Makefile ##
>      @@ Makefile: $(BUILT_INS): git$X
> 
> base-commit: 4a7958ca1415077ce9b1d0a38223ede55da779d9
> -- 
> 2.53.0
> 

> Le 23 févr. 2026 à 01:55, SZEDER Gábor <szeder.dev@gmail.com> a écrit :
> 
> On Sat, Feb 21, 2026 at 09:07:17AM -0500, D. Ben Knoble wrote:
>> The Meson-based build doesn't know when to rebuild config-list.h, so the
>> header is sometimes stale.
>> 
>> For example, an old build directory might have config-list.h from before
>> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
>> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
>> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
>> from sources fixes the artifact and the test.
>> 
>> Since Meson does not have (or want) builtin support for globbing like
>> Make, teach generate-configlist.sh to also generate a list of
>> Documentation files its output depends on, and incorporate that into the
>> Meson build.
>> 
>> We assume that if a user adds a new file under
>> Documentation/config then they will also edit one of the existing files
>> to include that new file, and that will trigger a rebuild. Also mark the
>> generator script as a dependency.
>> 
>> While we're at it, teach the Makefile to use the same "the script knows
>> it's dependencies" logic.
>> 
>> For Meson, combining the following commands helps debug dependencies:
>> 
>>    ninja -C <builddir> -t deps config-list.h
>>    ninja -C <builddir> -t browse config-list.h
>> 
>> The former lists all the dependencies discovered from our output ".d"
>> file (the config documentation) and the latter shows the dependency on
>> the script itself, among other useful edges in the dependency graph.
>> 
>> Helped-by: Patrick Steinhardt <ps@pks.im>
>> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
>> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
>> ---
>> 
>> Notes (benknoble/commits):
>>    Changes from v5 (<611a94cd988e3795bc63dba2f1b270aa0d058bd2.1771425395.git.ben.knoble+github@gmail.com>):
>> 
>>    • Reword a confusing sentence in the commit message
>> 
>> Makefile               |  5 +++--
>> generate-configlist.sh | 11 ++++++++++-
>> meson.build            |  5 ++++-
>> 3 files changed, 17 insertions(+), 4 deletions(-)
>> 
>> diff --git a/Makefile b/Makefile
>> index 7f37ad8f58..6f926ffb1f 100644
>> --- a/Makefile
>> +++ b/Makefile
>> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
>>    cp $< $@
>> 
>> config-list.h: generate-configlist.sh
>> +    @mkdir -p .depend
>> +    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>> 
>> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
>> -    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
>> +-include .depend/config-list.h.d
> 
> This breaks the build when something disappears from
> Documentation/config/:
> 
>  $ git checkout origin/seen
>  HEAD is now at 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
>  $ ls -l Documentation/config/hook.adoc
>  -rw-rw-r-- 1 szeder szeder 3828 Feb 23 07:50 Documentation/config/hook.adoc
>  $ git grep hook.adoc
>  Documentation/git-hook.adoc:include::config/hook.adoc[]
>  Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
>  Documentation/meson.build:  'git-hook.adoc' : 1,
>  $ make V=1 config-list.h
>  /bin/sh ./generate-configlist.sh . config-list.h .depend/config-list.h.d
>  $ git checkout 0aabf70f60
>  Previous HEAD position was 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
>  HEAD is now at 0aabf70f60 build: regenerate config-list.h when Documentation changes
>  $ ls -l Documentation/config/hook.adoc
>  ls: cannot access 'Documentation/config/hook.adoc': No such file or directory
>  $ git grep hook.adoc
>  Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
>  Documentation/meson.build:  'git-hook.adoc' : 1,
>  $ make V=1 config-list.h
>  GIT_VERSION=2.53.0.119.g0aabf70f60
>  make: *** No rule to make target 'Documentation/config/hook.adoc', needed by 'config-list.h'.  Stop.
>  $ grep hook.adoc .depend/config-list.h.d
>  config-list.h: ./Documentation/config/hook.adoc

> > Le 23 févr. 2026 à 01:55, SZEDER Gábor <szeder.dev@gmail.com> a écrit :
> > On Sat, Feb 21, 2026 at 09:07:17AM -0500, D. Ben Knoble wrote:
> >> diff --git a/Makefile b/Makefile
> >> index 7f37ad8f58..6f926ffb1f 100644
> >> --- a/Makefile
> >> +++ b/Makefile
> >> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
> >>    cp $< $@
> >> 
> >> config-list.h: generate-configlist.sh
> >> +    @mkdir -p .depend
> >> +    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
> >> 
> >> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
> >> -    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
> >> +-include .depend/config-list.h.d
> > 
> > This breaks the build when something disappears from
> > Documentation/config/:
> > 
> >  $ git checkout origin/seen
> >  HEAD is now at 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
> >  $ ls -l Documentation/config/hook.adoc
> >  -rw-rw-r-- 1 szeder szeder 3828 Feb 23 07:50 Documentation/config/hook.adoc
> >  $ git grep hook.adoc
> >  Documentation/git-hook.adoc:include::config/hook.adoc[]
> >  Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
> >  Documentation/meson.build:  'git-hook.adoc' : 1,
> >  $ make V=1 config-list.h
> >  /bin/sh ./generate-configlist.sh . config-list.h .depend/config-list.h.d
> >  $ git checkout 0aabf70f60
> >  Previous HEAD position was 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
> >  HEAD is now at 0aabf70f60 build: regenerate config-list.h when Documentation changes
> >  $ ls -l Documentation/config/hook.adoc
> >  ls: cannot access 'Documentation/config/hook.adoc': No such file or directory
> >  $ git grep hook.adoc
> >  Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
> >  Documentation/meson.build:  'git-hook.adoc' : 1,
> >  $ make V=1 config-list.h
> >  GIT_VERSION=2.53.0.119.g0aabf70f60
> >  make: *** No rule to make target 'Documentation/config/hook.adoc', needed by 'config-list.h'.  Stop.
> >  $ grep hook.adoc .depend/config-list.h.d
> >  config-list.h: ./Documentation/config/hook.adoc
> 
> Indeed. This might arise while bisecting, which was my original
> motivation. Thoughts on a path forward? At least this issue (to me) is
> clearer than a spurious test failure :)

> On Mon, Feb 23, 2026 at 04:41:48PM -0500, Ben Knoble wrote:
>>> Le 23 févr. 2026 à 01:55, SZEDER Gábor <szeder.dev@gmail.com> a écrit :
>>> On Sat, Feb 21, 2026 at 09:07:17AM -0500, D. Ben Knoble wrote:
>>>> diff --git a/Makefile b/Makefile
>>>> index 7f37ad8f58..6f926ffb1f 100644
>>>> --- a/Makefile
>>>> +++ b/Makefile
>>>> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
>>>>     cp $< $@
>>>>
>>>> config-list.h: generate-configlist.sh
>>>> +    @mkdir -p .depend
>>>> +    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
>>>>
>>>> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
>>>> -    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
>>>> +-include .depend/config-list.h.d
>>>
>>> This breaks the build when something disappears from
>>> Documentation/config/:
>>>
>>>   $ git checkout origin/seen
>>>   HEAD is now at 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
>>>   $ ls -l Documentation/config/hook.adoc
>>>   -rw-rw-r-- 1 szeder szeder 3828 Feb 23 07:50 Documentation/config/hook.adoc
>>>   $ git grep hook.adoc
>>>   Documentation/git-hook.adoc:include::config/hook.adoc[]
>>>   Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
>>>   Documentation/meson.build:  'git-hook.adoc' : 1,
>>>   $ make V=1 config-list.h
>>>   /bin/sh ./generate-configlist.sh . config-list.h .depend/config-list.h.d
>>>   $ git checkout 0aabf70f60
>>>   Previous HEAD position was 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
>>>   HEAD is now at 0aabf70f60 build: regenerate config-list.h when Documentation changes
>>>   $ ls -l Documentation/config/hook.adoc
>>>   ls: cannot access 'Documentation/config/hook.adoc': No such file or directory
>>>   $ git grep hook.adoc
>>>   Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
>>>   Documentation/meson.build:  'git-hook.adoc' : 1,
>>>   $ make V=1 config-list.h
>>>   GIT_VERSION=2.53.0.119.g0aabf70f60
>>>   make: *** No rule to make target 'Documentation/config/hook.adoc', needed by 'config-list.h'.  Stop.
>>>   $ grep hook.adoc .depend/config-list.h.d
>>>   config-list.h: ./Documentation/config/hook.adoc
>>
>> Indeed. This might arise while bisecting, which was my original
>> motivation. Thoughts on a path forward? At least this issue (to me) is
>> clearer than a spurious test failure :)
> 
> For Meson this case works alright. So maybe we just drop the changes to
> the Makefile and call it a day?
> 
> An alternative would be to have the following patch on top:
> 
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 39ac8845ab..e28054f9e0 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -41,7 +41,12 @@ EOF
>   if test -n "$DEPFILE"
>   then
>   	QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> -	printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> -		"$SOURCE_DIR"/Documentation/config/*.adoc |
> -		sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> +	{
> +		printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> +			"$SOURCE_DIR"/Documentation/config/*.adoc |
> +			sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /"
> +		printf '%s:\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> +			"$SOURCE_DIR"/Documentation/config/*.adoc |
> +			sed -e 's/[# ]/\\&/g'
> +	} >"$DEPFILE"
>   fi
> 
> What it does is to also create an empty target for all of the
> dependencies. Which is in fact what GCC/Clang also do when you pass -MP:

> 
>      $ cat main.c
>      #include "foo.h"
>      int main()
>      {
>          puts("foobar");
>          return 0;
>      }
> 
>      $ clang -MMD -MP main.c
>      $ cat main.d
>      main.o: main.c foo.h
>      foo.h:
> 
> Patrick

>
>
>
> On 24/02/2026 09:58, Patrick Steinhardt wrote:
> > On Mon, Feb 23, 2026 at 04:41:48PM -0500, Ben Knoble wrote:
> >>> Le 23 févr. 2026 à 01:55, SZEDER Gábor <szeder.dev@gmail.com> a écrit :
> >>> On Sat, Feb 21, 2026 at 09:07:17AM -0500, D. Ben Knoble wrote:
> >>>> diff --git a/Makefile b/Makefile
> >>>> index 7f37ad8f58..6f926ffb1f 100644
> >>>> --- a/Makefile
> >>>> +++ b/Makefile
> >>>> @@ -2688,9 +2688,10 @@ $(BUILT_INS): git$X
> >>>>     cp $< $@
> >>>>
> >>>> config-list.h: generate-configlist.sh
> >>>> +    @mkdir -p .depend
> >>>> +    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@ .depend/config-list.h.d
> >>>>
> >>>> -config-list.h: Documentation/*config.adoc Documentation/config/*.adoc
> >>>> -    $(QUIET_GEN)$(SHELL_PATH) ./generate-configlist.sh . $@
> >>>> +-include .depend/config-list.h.d
> >>>
> >>> This breaks the build when something disappears from
> >>> Documentation/config/:
> >>>
> >>>   $ git checkout origin/seen
> >>>   HEAD is now at 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
> >>>   $ ls -l Documentation/config/hook.adoc
> >>>   -rw-rw-r-- 1 szeder szeder 3828 Feb 23 07:50 Documentation/config/hook.adoc
> >>>   $ git grep hook.adoc
> >>>   Documentation/git-hook.adoc:include::config/hook.adoc[]
> >>>   Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
> >>>   Documentation/meson.build:  'git-hook.adoc' : 1,
> >>>   $ make V=1 config-list.h
> >>>   /bin/sh ./generate-configlist.sh . config-list.h .depend/config-list.h.d
> >>>   $ git checkout 0aabf70f60
> >>>   Previous HEAD position was 57edfa3ce8 Merge branch 'ty/setup-error-tightening' into seen
> >>>   HEAD is now at 0aabf70f60 build: regenerate config-list.h when Documentation changes
> >>>   $ ls -l Documentation/config/hook.adoc
> >>>   ls: cannot access 'Documentation/config/hook.adoc': No such file or directory
> >>>   $ git grep hook.adoc
> >>>   Documentation/howto/meson.build:  'rebuild-from-update-hook.adoc',
> >>>   Documentation/meson.build:  'git-hook.adoc' : 1,
> >>>   $ make V=1 config-list.h
> >>>   GIT_VERSION=2.53.0.119.g0aabf70f60
> >>>   make: *** No rule to make target 'Documentation/config/hook.adoc', needed by 'config-list.h'.  Stop.
> >>>   $ grep hook.adoc .depend/config-list.h.d
> >>>   config-list.h: ./Documentation/config/hook.adoc
> >>
> >> Indeed. This might arise while bisecting, which was my original
> >> motivation. Thoughts on a path forward? At least this issue (to me) is
> >> clearer than a spurious test failure :)
> >
> > For Meson this case works alright. So maybe we just drop the changes to
> > the Makefile and call it a day?
> >
> > An alternative would be to have the following patch on top:
> >
> > diff --git a/generate-configlist.sh b/generate-configlist.sh
> > index 39ac8845ab..e28054f9e0 100755
> > --- a/generate-configlist.sh
> > +++ b/generate-configlist.sh
> > @@ -41,7 +41,12 @@ EOF
> >   if test -n "$DEPFILE"
> >   then
> >       QUOTED_OUTPUT="$(printf '%s\n' "$OUTPUT" | sed 's,[&/\],\\&,g')"
> > -     printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> > -             "$SOURCE_DIR"/Documentation/config/*.adoc |
> > -             sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /" >"$DEPFILE"
> > +     {
> > +             printf '%s\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> > +                     "$SOURCE_DIR"/Documentation/config/*.adoc |
> > +                     sed -e 's/[# ]/\\&/g' -e "s/^/$QUOTED_OUTPUT: /"
> > +             printf '%s:\n' "$SOURCE_DIR"/Documentation/*config.adoc \
> > +                     "$SOURCE_DIR"/Documentation/config/*.adoc |
> > +                     sed -e 's/[# ]/\\&/g'
> > +     } >"$DEPFILE"
> >   fi
> >
> > What it does is to also create an empty target for all of the
> > dependencies. Which is in fact what GCC/Clang also do when you pass -MP:
>
> Oh, I was wondering how they got around the problem we're seeing here.
> While it seems ninja is happy it ignore missing dependencies in
> depfiles, are we sure the build systems meson uses on other platforms
> are as well? If not it might be worth adding this, rather than just
> dropping the Makefile changes.
>
> Thanks
>
> Phillip
>
> >
> >      $ cat main.c
> >      #include "foo.h"
> >      int main()
> >      {
> >          puts("foobar");
> >          return 0;
> >      }
> >
> >      $ clang -MMD -MP main.c
> >      $ cat main.d
> >      main.o: main.c foo.h
> >      foo.h:
> >
> > Patrick

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> ...
> Helped-by: Patrick Steinhardt <ps@pks.im>
> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
>
> Notes (benknoble/commits):
>     Changes from v6 (<5dcd4e9308100a25603c50fecb36447c0ee4df62.1771682788.git.ben.knoble+github@gmail.com>):
>     
>     • Fix Make-based builds when a dependency is removed by generating the
>       expected empty targets. (Thanks to a report from Szeder Gábor.)

> Le 25 févr. 2026 à 13:45, Junio C Hamano <gitster@pobox.com> a écrit :
> 
> "D. Ben Knoble" <ben.knoble+github@gmail.com> writes:
> 
>> The Meson-based build doesn't know when to rebuild config-list.h, so the
>> header is sometimes stale.
>> ...
>> Helped-by: Patrick Steinhardt <ps@pks.im>
>> Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
>> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
>> ---
>> 
>> Notes (benknoble/commits):
>>    Changes from v6 (<5dcd4e9308100a25603c50fecb36447c0ee4df62.1771682788.git.ben.knoble+github@gmail.com>):
>> 
>>    • Fix Make-based builds when a dependency is removed by generating the
>>      expected empty targets. (Thanks to a report from Szeder Gábor.)
> 
> OK, now we are pretty much completed and ready to merge it down to
> 'next'?  Let me mark it as such.
> 
> Thanks.

> The Meson-based build doesn't know when to rebuild config-list.h, so the
> header is sometimes stale.
> 
> For example, an old build directory might have config-list.h from before
> 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> from sources fixes the artifact and the test.
> 
> Teach the meson build to depend on the Documentation files that
> generate-configlist.sh reads by having it use the output of
> generate-configlist-deps.sh as a list of dependency files, since Meson
> does not have (or want) builtin support for globbing like Make.
> 
> Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> ---
> 
> Notes (benknoble/commits):
>     I considered having generate-configlist.sh write its own dependency
>     list, which Meson also supports… idk though. Input welcome :)

> On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> > The Meson-based build doesn't know when to rebuild config-list.h, so the
> > header is sometimes stale.
> >
> > For example, an old build directory might have config-list.h from before
> > 4173df5187 (submodule: introduce extensions.submodulePathConfig,
> > 2026-01-12), which added submodule.<name>.gitdir to the list. Without
> > it, t9902-completion.sh fails. Regenerating the config-list.h artifact
> > from sources fixes the artifact and the test.
> >
> > Teach the meson build to depend on the Documentation files that
> > generate-configlist.sh reads by having it use the output of
> > generate-configlist-deps.sh as a list of dependency files, since Meson
> > does not have (or want) builtin support for globbing like Make.
> >
> > Signed-off-by: D. Ben Knoble <ben.knoble+github@gmail.com>
> > ---
> >
> > Notes (benknoble/commits):
> >     I considered having generate-configlist.sh write its own dependency
> >     list, which Meson also supports… idk though. Input welcome :)
>
> I think that would actually be the better approach, also because the
> list of files with `run_command()` would only be computed at setup time.

> I guess it could look something like the below patch -- please feel free
> to reuse it at will.
>
> Thanks!
>
> Patrick

> Only, things are behaving oddly. For example:
> 
>     λ meson setup build2
>     λ ninja -C build2
> 
> works fine, but
> 
>     λ ls -l build2/config*
>     -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
> 
> I don't see the dependency file.

> Further, re-building seems to get stuck (I get
> similar symptoms if I add or remove a relevant config.adoc file, but let's keep
> it simple for now):
> 
>     λ ninja -C build2
>     ninja: Entering directory `build2'
>     [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)

> On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
>>> On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> [snip]
>> Only, things are behaving oddly. For example:
>>
>>      λ meson setup build2
>>      λ ninja -C build2
>>
>> works fine, but
>>
>>      λ ls -l build2/config*
>>      -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
>>
>> I don't see the dependency file.
> 
>> Further, re-building seems to get stuck (I get
>> similar symptoms if I add or remove a relevant config.adoc file, but let's keep
>> it simple for now):
>>
>>      λ ninja -C build2
>>      ninja: Entering directory `build2'
>>      [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)
> 
> With "stuck" you mean that it doesn't do anything, or that it doesn't
> actually rebuild?
> 
> I guess it kind of makes sense that a new file wouldn't trigger a
> rebuild, even though I would have expected a removed one to trigger one.
> After all, the dependency file only tracks the set of _existing_ files
> so that we know when to rebuild, and of course the dependency file only
> gets regenerated in case any of those files changes.

> The thing is that build systems like Meson really want to know the list
> of files ahead of time so that they can have an optimal build graph. So
> we could of course list all the files that we actually depend on. But I
> guess that's something we want to avoid?
> 
> There's another, alternative approach: you can have a separate build
> step that's marked as `build_always_stale: true` that lists all the
> config files. This step would then always run, and it would only update
> its target file in case any of the files has changed.

> Combined with the depfile we'd then rebuild in all cases:
> 
>    - When a file gets added or removed, as that would cause the
>      `build_always_stale` target to be rewritten.
> 
>    - When any of the files changes, because that would cause the
>      dependencies in the depfile to change.
> 
> Something like the attached patch (note that I don't perform the
> necessary changes for the Makefile). I confirmed that it works for all
> of the above cases.
> 
> Thanks!
> 
> Patrick
> 
> --- >8 ---
> 
> diff --git a/generate-configlist.sh b/generate-configlist.sh
> index 75c39ade20..17605e6f77 100755
> --- a/generate-configlist.sh
> +++ b/generate-configlist.sh
> @@ -1,13 +1,6 @@
>   #!/bin/sh
>   
> -SOURCE_DIR="$1"
> -OUTPUT="$2"
> -
> -if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
> -then
> -	echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
> -	exit 1
> -fi
> +set -e
>   
>   print_config_list () {
>   	cat <<EOF
> @@ -30,9 +23,50 @@ EOF
>   EOF
>   }
>   
> -{
> -	echo "/* Automatically generated by generate-configlist.sh */"
> -	echo
> -	echo
> -	print_config_list
> -} >"$OUTPUT"
> +case "$1" in
> +generate)
> +	SOURCE_DIR="$2"
> +	OUTPUT="$3"
> +	DEPFILE="$4"
> +
> +	if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
> +	then
> +		echo >&2 "USAGE: $0 generate <SOURCE_DIR> <OUTPUT>"
> +		exit 1
> +	fi
> +
> +	if test -n "$DEPFILE"
> +	then
> +		printf "$OUTPUT: %s\n" "$SOURCE_DIR"/Documentation/*config.adoc \
> +			"$SOURCE_DIR"/Documentation/config/*.adoc >"$DEPFILE"
> +	fi
> +
> +	{
> +		echo "/* Automatically generated by generate-configlist.sh */"
> +		echo
> +		echo
> +		print_config_list
> +	} >"$OUTPUT"
> +	;;
> +deps)
> +	SOURCE_DIR="$2"
> +	OUTPUT="$3"
> +
> +	if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
> +	then
> +		echo >&2 "USAGE: $0 deps <SOURCE_DIR> <OUTPUT>"
> +		exit 1
> +	fi
> +
> +	TMPFILE=$(mktemp "$OUTPUT".XXXXXX)
> +	printf "%s\n" "$SOURCE_DIR"/Documentation/*config.adoc \
> +		"$SOURCE_DIR"/Documentation/config/*.adoc | sort >"$TMPFILE"
> +
> +	if ! test -f "$OUTPUT" || ! cmp "$TMPFILE" "$OUTPUT" >/dev/null
> +	then
> +		mv "$TMPFILE" "$OUTPUT"
> +	else
> +		rm "$TMPFILE"
> +	fi
> +	;;
> +esac
> diff --git a/meson.build b/meson.build
> index dd52efd1c8..6b9147a39a 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -716,14 +716,29 @@ if not get_option('breaking_changes')
>     builtin_sources += 'builtin/pack-redundant.c'
>   endif
>   
> +configlist_deps = custom_target(
> +  output: 'config-list.h.deps',
> +  command: [
> +    meson.current_source_dir() + '/generate-configlist.sh',
> +    'deps',
> +    meson.current_source_dir(),
> +    '@OUTPUT@',
> +  ],
> +  build_always_stale: true,
> +)
> +
>   builtin_sources += custom_target(
>     output: 'config-list.h',
>     command: [
>       shell,
>       meson.current_source_dir() + '/generate-configlist.sh',
> +    'generate',
>       meson.current_source_dir(),
>       '@OUTPUT@',
> +    '@OUTPUT@.d',
>     ],
> +  depends: [ configlist_deps, ],
> +  depfile: 'config-list.h.d',
>     env: script_environment,
>   )
>   
> 

> On 11/02/2026 07:42, Patrick Steinhardt wrote:
>> On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
>>>> On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
>> [snip]
>>> Only, things are behaving oddly. For example:
>>>
>>>      λ meson setup build2
>>>      λ ninja -C build2
>>>
>>> works fine, but
>>>
>>>      λ ls -l build2/config*
>>>      -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/ 
>>> config-list.h
>>>
>>> I don't see the dependency file.

>>> Further, re-building seems to get stuck (I get
>>> similar symptoms if I add or remove a relevant config.adoc file, but 
>>> let's keep
>>> it simple for now):
>>>
>>>      λ ninja -C build2
>>>      ninja: Entering directory `build2'
>>>      [1/28] Generating GIT-VERSION-FILE with a custom command 
>>> (wrapped by meson to set env)
>>
>> With "stuck" you mean that it doesn't do anything, or that it doesn't
>> actually rebuild?
>>
>> I guess it kind of makes sense that a new file wouldn't trigger a
>> rebuild, even though I would have expected a removed one to trigger one.
>> After all, the dependency file only tracks the set of _existing_ files
>> so that we know when to rebuild, and of course the dependency file only
>> gets regenerated in case any of those files changes.
> 
> If anyone adds a new file under Documentation/config/ they will need to 
> update Documentation/config.adoc which should then trigger the rebuild. 
> That rebuld will then add the new file to the list of dependencies. If 
> they remove a file we should pick that up with the dependencies that are 
> already listed.
> 
> Do we need to create the depfile when meson is setup? What does meson do 
> when the depfile is missing?
> 
>> The thing is that build systems like Meson really want to know the list
>> of files ahead of time so that they can have an optimal build graph. So
>> we could of course list all the files that we actually depend on. But I
>> guess that's something we want to avoid?
>>
>> There's another, alternative approach: you can have a separate build
>> step that's marked as `build_always_stale: true` that lists all the
>> config files. This step would then always run, and it would only update
>> its target file in case any of the files has changed.
> 
> It would be really nice if we can avoid regenerating the depfile with 
> every build.
> 
> Thanks
> 
> Phillip
> 
>> Combined with the depfile we'd then rebuild in all cases:
>>
>>    - When a file gets added or removed, as that would cause the
>>      `build_always_stale` target to be rewritten.
>>
>>    - When any of the files changes, because that would cause the
>>      dependencies in the depfile to change.
>>
>> Something like the attached patch (note that I don't perform the
>> necessary changes for the Makefile). I confirmed that it works for all
>> of the above cases.
>>
>> Thanks!
>>
>> Patrick
>>
>> --- >8 ---
>>
>> diff --git a/generate-configlist.sh b/generate-configlist.sh
>> index 75c39ade20..17605e6f77 100755
>> --- a/generate-configlist.sh
>> +++ b/generate-configlist.sh
>> @@ -1,13 +1,6 @@
>>   #!/bin/sh
>> -SOURCE_DIR="$1"
>> -OUTPUT="$2"
>> -
>> -if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z "$OUTPUT"
>> -then
>> -    echo >&2 "USAGE: $0 <SOURCE_DIR> <OUTPUT>"
>> -    exit 1
>> -fi
>> +set -e
>>   print_config_list () {
>>       cat <<EOF
>> @@ -30,9 +23,50 @@ EOF
>>   EOF
>>   }
>> -{
>> -    echo "/* Automatically generated by generate-configlist.sh */"
>> -    echo
>> -    echo
>> -    print_config_list
>> -} >"$OUTPUT"
>> +case "$1" in
>> +generate)
>> +    SOURCE_DIR="$2"
>> +    OUTPUT="$3"
>> +    DEPFILE="$4"
>> +
>> +    if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z 
>> "$OUTPUT"
>> +    then
>> +        echo >&2 "USAGE: $0 generate <SOURCE_DIR> <OUTPUT>"
>> +        exit 1
>> +    fi
>> +
>> +    if test -n "$DEPFILE"
>> +    then
>> +        printf "$OUTPUT: %s\n" "$SOURCE_DIR"/Documentation/ 
>> *config.adoc \
>> +            "$SOURCE_DIR"/Documentation/config/*.adoc >"$DEPFILE"
>> +    fi
>> +
>> +    {
>> +        echo "/* Automatically generated by generate-configlist.sh */"
>> +        echo
>> +        echo
>> +        print_config_list
>> +    } >"$OUTPUT"
>> +    ;;
>> +deps)
>> +    SOURCE_DIR="$2"
>> +    OUTPUT="$3"
>> +
>> +    if test -z "$SOURCE_DIR" || ! test -d "$SOURCE_DIR" || test -z 
>> "$OUTPUT"
>> +    then
>> +        echo >&2 "USAGE: $0 deps <SOURCE_DIR> <OUTPUT>"
>> +        exit 1
>> +    fi
>> +
>> +    TMPFILE=$(mktemp "$OUTPUT".XXXXXX)
>> +    printf "%s\n" "$SOURCE_DIR"/Documentation/*config.adoc \
>> +        "$SOURCE_DIR"/Documentation/config/*.adoc | sort >"$TMPFILE"
>> +
>> +    if ! test -f "$OUTPUT" || ! cmp "$TMPFILE" "$OUTPUT" >/dev/null
>> +    then
>> +        mv "$TMPFILE" "$OUTPUT"
>> +    else
>> +        rm "$TMPFILE"
>> +    fi
>> +    ;;
>> +esac
>> diff --git a/meson.build b/meson.build
>> index dd52efd1c8..6b9147a39a 100644
>> --- a/meson.build
>> +++ b/meson.build
>> @@ -716,14 +716,29 @@ if not get_option('breaking_changes')
>>     builtin_sources += 'builtin/pack-redundant.c'
>>   endif
>> +configlist_deps = custom_target(
>> +  output: 'config-list.h.deps',
>> +  command: [
>> +    meson.current_source_dir() + '/generate-configlist.sh',
>> +    'deps',
>> +    meson.current_source_dir(),
>> +    '@OUTPUT@',
>> +  ],
>> +  build_always_stale: true,
>> +)
>> +
>>   builtin_sources += custom_target(
>>     output: 'config-list.h',
>>     command: [
>>       shell,
>>       meson.current_source_dir() + '/generate-configlist.sh',
>> +    'generate',
>>       meson.current_source_dir(),
>>       '@OUTPUT@',
>> +    '@OUTPUT@.d',
>>     ],
>> +  depends: [ configlist_deps, ],
>> +  depfile: 'config-list.h.d',
>>     env: script_environment,
>>   )
>>
> 

> On 11/02/2026 09:44, Phillip Wood wrote:
> > On 11/02/2026 07:42, Patrick Steinhardt wrote:
> > > On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
> > > > > On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> > > [snip]
> > > > Only, things are behaving oddly. For example:
> > > > 
> > > >      λ meson setup build2
> > > >      λ ninja -C build2
> > > > 
> > > > works fine, but
> > > > 
> > > >      λ ls -l build2/config*
> > > >      -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39
> > > > build2/ config-list.h
> > > > 
> > > > I don't see the dependency file.
> 
> I've just had a play with this and I don't see the dependency file either -
> I wonder if ninja deletes it after it has read it.

> On 11/02/2026 07:42, Patrick Steinhardt wrote:
> > On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
> > > > On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> > [snip]
> > > Only, things are behaving oddly. For example:
> > > 
> > >      λ meson setup build2
> > >      λ ninja -C build2
> > > 
> > > works fine, but
> > > 
> > >      λ ls -l build2/config*
> > >      -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
> > > 
> > > I don't see the dependency file.
> > 
> > > Further, re-building seems to get stuck (I get
> > > similar symptoms if I add or remove a relevant config.adoc file, but let's keep
> > > it simple for now):
> > > 
> > >      λ ninja -C build2
> > >      ninja: Entering directory `build2'
> > >      [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)
> > 
> > With "stuck" you mean that it doesn't do anything, or that it doesn't
> > actually rebuild?
> > 
> > I guess it kind of makes sense that a new file wouldn't trigger a
> > rebuild, even though I would have expected a removed one to trigger one.
> > After all, the dependency file only tracks the set of _existing_ files
> > so that we know when to rebuild, and of course the dependency file only
> > gets regenerated in case any of those files changes.
> 
> If anyone adds a new file under Documentation/config/ they will need to
> update Documentation/config.adoc which should then trigger the rebuild. That
> rebuld will then add the new file to the list of dependencies. If they
> remove a file we should pick that up with the dependencies that are already
> listed.

> > The thing is that build systems like Meson really want to know the list
> > of files ahead of time so that they can have an optimal build graph. So
> > we could of course list all the files that we actually depend on. But I
> > guess that's something we want to avoid?
> > 
> > There's another, alternative approach: you can have a separate build
> > step that's marked as `build_always_stale: true` that lists all the
> > config files. This step would then always run, and it would only update
> > its target file in case any of the files has changed.
> 
> It would be really nice if we can avoid regenerating the depfile with every
> build.

> On Wed, Feb 11, 2026 at 09:44:48AM +0000, Phillip Wood wrote:
>> On 11/02/2026 07:42, Patrick Steinhardt wrote:
>>> On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
>>>>> On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
>>> [snip]
>>>> Only, things are behaving oddly. For example:
>>>>
>>>>       λ meson setup build2
>>>>       λ ninja -C build2
>>>>
>>>> works fine, but
>>>>
>>>>       λ ls -l build2/config*
>>>>       -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
>>>>
>>>> I don't see the dependency file.
>>>
>>>> Further, re-building seems to get stuck (I get
>>>> similar symptoms if I add or remove a relevant config.adoc file, but let's keep
>>>> it simple for now):
>>>>
>>>>       λ ninja -C build2
>>>>       ninja: Entering directory `build2'
>>>>       [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)
>>>
>>> With "stuck" you mean that it doesn't do anything, or that it doesn't
>>> actually rebuild?
>>>
>>> I guess it kind of makes sense that a new file wouldn't trigger a
>>> rebuild, even though I would have expected a removed one to trigger one.
>>> After all, the dependency file only tracks the set of _existing_ files
>>> so that we know when to rebuild, and of course the dependency file only
>>> gets regenerated in case any of those files changes.
>>
>> If anyone adds a new file under Documentation/config/ they will need to
>> update Documentation/config.adoc which should then trigger the rebuild. That
>> rebuld will then add the new file to the list of dependencies. If they
>> remove a file we should pick that up with the dependencies that are already
>> listed.
> 
> Oh? Well, if that's the case then the additional changes should indeed
> not be required.

>>
>> It would be really nice if we can avoid regenerating the depfile with every
>> build.
> 
> Agreed. So maybe the first patch I sent is sufficient after all?

>
> On 11/02/2026 10:58, Patrick Steinhardt wrote:
> > On Wed, Feb 11, 2026 at 09:44:48AM +0000, Phillip Wood wrote:
> >> On 11/02/2026 07:42, Patrick Steinhardt wrote:
> >>> On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
> >>>>> On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> >>> [snip]
> >>>> Only, things are behaving oddly. For example:
> >>>>
> >>>>       λ meson setup build2
> >>>>       λ ninja -C build2
> >>>>
> >>>> works fine, but
> >>>>
> >>>>       λ ls -l build2/config*
> >>>>       -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
> >>>>
> >>>> I don't see the dependency file.
> >>>
> >>>> Further, re-building seems to get stuck (I get
> >>>> similar symptoms if I add or remove a relevant config.adoc file, but let's keep
> >>>> it simple for now):
> >>>>
> >>>>       λ ninja -C build2
> >>>>       ninja: Entering directory `build2'
> >>>>       [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)
> >>>
> >>> With "stuck" you mean that it doesn't do anything, or that it doesn't
> >>> actually rebuild?
> >>>
> >>> I guess it kind of makes sense that a new file wouldn't trigger a
> >>> rebuild, even though I would have expected a removed one to trigger one.
> >>> After all, the dependency file only tracks the set of _existing_ files
> >>> so that we know when to rebuild, and of course the dependency file only
> >>> gets regenerated in case any of those files changes.
> >>
> >> If anyone adds a new file under Documentation/config/ they will need to
> >> update Documentation/config.adoc which should then trigger the rebuild. That
> >> rebuld will then add the new file to the list of dependencies. If they
> >> remove a file we should pick that up with the dependencies that are already
> >> listed.
> >
> > Oh? Well, if that's the case then the additional changes should indeed
> > not be required.
>
> I think so. The only problem I can think of is that if you delete a
> file, build, restore the file without changing anything else and build
> again then config-list.h will not be rebuilt because the deleted file
> would have been removed from the list of dependencies by the previous build.
>
> >>
> >> It would be really nice if we can avoid regenerating the depfile with every
> >> build.
> >
> > Agreed. So maybe the first patch I sent is sufficient after all?
>
> We need to add the script to the list of dependencies and reading the
> comments in
> https://github.com/ninja-build/ninja/blob/master/src/depfile_parser.in.cc
> we should be backslash escaping space, hash and backslash in the
> filename when we write the dependencies. Apart from that I think it is ok.

> I guess the alternative is to bite the bullet and list these
> dependencies explicitly as we do for other targets.
>
> Thanks
>
> Phillip
>

>
> On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
> > > On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> [snip]
> > Only, things are behaving oddly. For example:
> >
> >     λ meson setup build2
> >     λ ninja -C build2
> >
> > works fine, but
> >
> >     λ ls -l build2/config*
> >     -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
> >
> > I don't see the dependency file.
>
> > Further, re-building seems to get stuck (I get
> > similar symptoms if I add or remove a relevant config.adoc file, but let's keep
> > it simple for now):
> >
> >     λ ninja -C build2
> >     ninja: Entering directory `build2'
> >     [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)
>
> With "stuck" you mean that it doesn't do anything, or that it doesn't
> actually rebuild?

> I guess it kind of makes sense that a new file wouldn't trigger a
> rebuild, even though I would have expected a removed one to trigger one.
> After all, the dependency file only tracks the set of _existing_ files
> so that we know when to rebuild, and of course the dependency file only
> gets regenerated in case any of those files changes.

> I'm not totally sure where the discussion ended up (what's actionable
> isn't clear to me), but let me respond to at least this bit:
> 
> On Wed, Feb 11, 2026 at 2:43 AM Patrick Steinhardt <ps@pks.im> wrote:
> >
> > On Mon, Feb 09, 2026 at 04:50:06PM -0500, D. Ben Knoble wrote:
> > > > On Sat, Feb 07, 2026 at 04:59:17PM -0500, D. Ben Knoble wrote:
> > [snip]
> > > Only, things are behaving oddly. For example:
> > >
> > >     λ meson setup build2
> > >     λ ninja -C build2
> > >
> > > works fine, but
> > >
> > >     λ ls -l build2/config*
> > >     -rw-r--r-- 1 benknoble benknoble 17169  9 févr. 16:39 build2/config-list.h
> > >
> > > I don't see the dependency file.
> >
> > > Further, re-building seems to get stuck (I get
> > > similar symptoms if I add or remove a relevant config.adoc file, but let's keep
> > > it simple for now):
> > >
> > >     λ ninja -C build2
> > >     ninja: Entering directory `build2'
> > >     [1/28] Generating GIT-VERSION-FILE with a custom command (wrapped by meson to set env)
> >
> > With "stuck" you mean that it doesn't do anything, or that it doesn't
> > actually rebuild?
> 
> I mean, that's all the output I get, and the build seems unable to
> make progress without the depfile present. Maybe I'm misreading the
> above output, so let me try again with a different kind of change that
> would cause recompilation:

> > I guess it kind of makes sense that a new file wouldn't trigger a
> > rebuild, even though I would have expected a removed one to trigger one.
> > After all, the dependency file only tracks the set of _existing_ files
> > so that we know when to rebuild, and of course the dependency file only
> > gets regenerated in case any of those files changes.
> 
> Trying this bit again, yeah, config-list.h doesn't get rebuilt if I do
> `touch Documentation/fooconfig.adoc` (I think make would rebuild
> though). And I can't tell if that's due to the weird [1/28] output or
> due to something else, since I can't "see" if Meson recomputed the
> depfile or not; with the depfile gone, it seems like there's nothing
> present to tell Meson (ninja) when to rebuild, which is weird to me.
> It seems this is expected, but it makes it difficult for me to tell
> what's going on here.
> 
> After `touch Documentation/config.adoc`, ninja executes 7 steps,
> including rebuilding the config-list.h, at which point it also knows
> to rebuild after touching "fooconfig.adoc" again. So maybe that's
> expected?