Hi brian,
On Fri, Aug 09, 2019 at 10:02:17AM +0000, brian m. carlson wrote:
> When applying multiple patches with git am, or when rebasing using the
> am backend, it's possible that one of our patches has updated a
> gitattributes file. Currently, we cache this information, so if a
> file in a subsequent patch has attributes applied, the file will be
> written out with the attributes in place as of the time we started the
> rebase or am operation, not with the attributes applied by the previous
> patch. This problem does not occur when using the -m or -i flags to
> rebase.
Ah. Thanks for a description of the issue.
> To ensure we write the correct data into the working tree, expire the
> cache after each patch that touches a path ending in ".gitattributes".
OK, that seems like a sensible direction...
> Signed-off-by: brian m. carlson <sandals@xxxxxxxxxxxxxxxxxxxx>
> ---
> apply.c | 11 +++++++++++
> convert.c | 9 ++++++++-
> convert.h | 6 ++++++
> t/t3400-rebase.sh | 23 +++++++++++++++++++++++
> 4 files changed, 48 insertions(+), 1 deletion(-)
>
> diff --git a/apply.c b/apply.c
> index cde95369bb..b959b88b8e 100644
> --- a/apply.c
> +++ b/apply.c
> @@ -4643,6 +4643,7 @@ static int apply_patch(struct apply_state *state,
> struct patch *list = NULL, **listp = &list;
> int skipped_patch = 0;
> int res = 0;
> + int flush_attributes = 0;
>
> state->patch_input_file = filename;
> if (read_patch_file(&buf, fd) < 0)
> @@ -4670,6 +4671,14 @@ static int apply_patch(struct apply_state *state,
> patch_stats(state, patch);
> *listp = patch;
> listp = &patch->next;
> +
> + if (!flush_attributes && patch->new_name) {
> + char *dummy = strip_path_suffix(patch->new_name, GITATTRIBUTES_FILE);
It's a shame that 'strip_path_suffix' doesn't take a 'char *out', and
accept NULL for that (which would save us the assignment and subsequent
'free'). In either case, this is certainly the appropriate function.
But, I'm not sure if 'dummy' is the best name for this variable or not.
My first thought was that I'd expect this to be named 'suffix' (or the
less descriptive 'p'), but I don't know if the change is warranted.
What *are* we supposed to call this variable?
"path_after_gitattributes" ;-)?
> + /* The patch applied to a .gitattributes file. */
> + if (dummy)
> + flush_attributes = 1;
> + free(dummy);
> + }
> }
> else {
> if (state->apply_verbosity > verbosity_normal)
> @@ -4746,6 +4755,8 @@ static int apply_patch(struct apply_state *state,
> if (state->summary && state->apply_verbosity > verbosity_silent)
> summary_patch_list(list);
>
> + if (flush_attributes)
> + reset_parsed_attributes();
> end:
> free_patch_list(list);
> strbuf_release(&buf);
> diff --git a/convert.c b/convert.c
> index 94ff837649..030e9b81b9 100644
> --- a/convert.c
> +++ b/convert.c
> @@ -1293,10 +1293,11 @@ struct conv_attrs {
> const char *working_tree_encoding; /* Supported encoding or default encoding if NULL */
> };
>
> +static struct attr_check *check;
> +
> static void convert_attrs(const struct index_state *istate,
> struct conv_attrs *ca, const char *path)
> {
> - static struct attr_check *check;
> struct attr_check_item *ccheck = NULL;
>
> if (!check) {
> @@ -1339,6 +1340,12 @@ static void convert_attrs(const struct index_state *istate,
> ca->crlf_action = CRLF_AUTO_INPUT;
> }
>
> +void reset_parsed_attributes(void)
> +{
> + attr_check_free(check);
> + check = NULL;
> +}
> +
Makes sense.
> int would_convert_to_git_filter_fd(const struct index_state *istate, const char *path)
> {
> struct conv_attrs ca;
> diff --git a/convert.h b/convert.h
> index 831559f10d..3710969d43 100644
> --- a/convert.h
> +++ b/convert.h
> @@ -94,6 +94,12 @@ void convert_to_git_filter_fd(const struct index_state *istate,
> int would_convert_to_git_filter_fd(const struct index_state *istate,
> const char *path);
>
> +/*
> + * Reset the internal list of attributes used by convert_to_git and
> + * convert_to_working_tree.
> + */
> +void reset_parsed_attributes(void);
> +
> /*****************************************************************
> *
> * Streaming conversion support
> diff --git a/t/t3400-rebase.sh b/t/t3400-rebase.sh
> index 80b23fd326..062dc41df7 100755
> --- a/t/t3400-rebase.sh
> +++ b/t/t3400-rebase.sh
> @@ -301,6 +301,29 @@ test_expect_success 'rebase --am and --show-current-patch' '
> )
> '
>
> +test_expect_success 'rebase --am and .gitattributes' '
> + test_create_repo attributes &&
> + (
> + cd attributes &&
> + test_commit init &&
> + test_config filter.test.clean "sed -e '\''s/smudged/clean/g'\''" &&
> + test_config filter.test.smudge "sed -e '\''s/clean/smudged/g'\''" &&
> +
> + test_commit second &&
> + git checkout -b test HEAD^ &&
> +
> + echo "*.txt filter=test" >.gitattributes &&
> + git add .gitattributes &&
> + test_commit third &&
> +
> + echo "This text is smudged." >a.txt &&
> + git add a.txt &&
> + test_commit fourth &&
> + git rebase master &&
> + grep "smudged" a.txt
> + )
> +'
> +
This definitely exercise the behavior here. Thanks for adding a test
while you're at it.
> test_expect_success 'rebase--merge.sh and --show-current-patch' '
> test_create_repo conflict-merge &&
> (
I wouldn't be opposed to renaming 'dummy' to be something else, but in
the case that you don't feel a re-roll is necessary, please consider my:
Reviewed-by: Taylor Blau <me@xxxxxxxxxxxx>
Thanks,
Taylor
