[PATCH v6 00/10] Make submodules work if .gitmodules is not checked out
- Date: Fri, 5 Oct 2018 15:05:51 +0200
- From: Antonio Ospite <ao2@xxxxxx>
- Subject: [PATCH v6 00/10] Make submodules work if .gitmodules is not checked out
this series teaches git to try and read the .gitmodules file from the
index (:.gitmodules) or from the current branch (HEAD:.gitmodules) when
the file is not readily available in the working tree.
This can be used, together with sparse checkouts, to enable submodule
usage with programs like vcsh which manage multiple repositories with
their working trees sharing the same path.
Thanks to SZEDER Gábor we found out that the changes in patch 9 could
allow to access the object store in an inconsistent way when using
multi-threading in "git grep --recurse-submodules", this has been dealt
with in this revision.
BTW, with Stefan Beller we also identified some unneeded code which
could have been removed to alleviate the issue, but that would not have
solved it completely; so, I am not removing the unnecessary call to
repo_read_gitmodules() builtin/grep.c in this series, possibly this can
become a stand-alone change.
The problems from above also made me investigate what implications the
current use of a global object store had on my new addition, and now
I know that there is one case which I cannot support just yet: nested
submodules without .gitmodules in their working tree.
This case has been documented with a warning and test_expect_failure
items in tests, and hitting the unsupported case does not alter the
current behavior of git.
Apart form patch 9 and 10 there are no major changes to the previous
IMHO we are in a place where the problem has been analyzed with enough
depth, the limitations have been identified and dealt with in a way that
should not affect current users nor diminish the usefulness of the new
v5 of the series is here:
v4 of the series is here:
v3 of the series is here:
v2 of the series is here:
v1 of the series, with some background info, is here:
Changes since v5:
* print_config_from_gitmodules() in patch 1 now accepts a struct
* submodule--helper config in patch 5 has been adjusted to use the new
signature of print_config_from_gitmodules().
* In patch 9 the grep read lock in builtin/grep.c now covers all code
paths involving config_from_gitmodules().
FTR git-grep is the only place where config_from_gitmodules() is
called from multi-threaded code.
* Patch 9 also documents the rare case that cannot be supported just
yet, and adds a warning to the user.
* In patch 9, config_from_gitmodules() now does not read any config
when the config source is not specified.(I added a catch-all "else"
block) This match more closely the behavior of the old code using
* Added a new test tool in patch 10 to exercise config_read_config()
in a more direct way, passing an arbitrary repository.
Admittedly, patch 10 performs a similar test to the one added to
t7814 in patch 9, so I'd be OK with dropping patch 10 if it is too
Antonio Ospite (10):
submodule: add a print_config_from_gitmodules() helper
submodule: factor out a config_set_in_gitmodules_file_gently function
t7411: merge tests 5 and 6
t7411: be nicer to future tests and really clean things up
submodule--helper: add a new 'config' subcommand
submodule: use the 'submodule--helper config' command
t7506: clean up .gitmodules properly before setting up new scenario
submodule: add a helper to check if it is safe to write to .gitmodules
submodule: support reading .gitmodules when it's not in the working
t/helper: add test-submodule-nested-repo-config
Makefile | 1 +
builtin/grep.c | 17 ++-
builtin/submodule--helper.c | 40 ++++++
cache.h | 2 +
git-submodule.sh | 13 +-
submodule-config.c | 68 ++++++++-
submodule-config.h | 2 +
submodule.c | 28 +++-
submodule.h | 1 +
t/helper/test-submodule-nested-repo-config.c | 30 ++++
t/helper/test-tool.c | 1 +
t/helper/test-tool.h | 1 +
t/t7411-submodule-config.sh | 141 +++++++++++++++++--
t/t7416-submodule-sparse-gitmodules.sh | 78 ++++++++++
t/t7506-status-submodule.sh | 3 +-
t/t7814-grep-recurse-submodules.sh | 16 +++
16 files changed, 410 insertions(+), 32 deletions(-)
create mode 100644 t/helper/test-submodule-nested-repo-config.c
create mode 100755 t/t7416-submodule-sparse-gitmodules.sh
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?