On Mon, Nov 23, 2020 at 04:05:05PM +0000, Derrick Stolee via GitGitGadget wrote:
>
>
> The git_config_set_multivar_in_file_gently() and related methods now
> take a 'flags' bitfield, so add a new bit representing the --fixed-value
> option from 'git config'. This alters the purpose of the value_regex
> parameter to be an exact string match. This requires some initialization
> changes in git_config_set_multivar_in_file_gently() and a new strcmp()
> call in the matches() method.
>
> The new CONFIG_FLAGS_FIXED_VALUE flag is initialized in builtin/config.c
> based on the --fixed-value option, and that needs to be updated in
> several callers.
>
> This patch only affects some of the modes of 'git config', and the rest
> will be completed in the next change.
>
> Signed-off-by: Derrick Stolee <dstolee@xxxxxxxxxxxxx>
> ---
> builtin/config.c | 16 ++++++++-----
> config.c | 5 +++++
> config.h | 7 ++++++
> t/t1300-config.sh | 57 +++++++++++++++++++++++++++++++++++++++++++++++
> 4 files changed, 80 insertions(+), 5 deletions(-)
>
> diff --git a/builtin/config.c b/builtin/config.c
> index bfb55a96df..3e49e04411 100644
> --- a/builtin/config.c
> +++ b/builtin/config.c
> @@ -616,6 +616,7 @@ int cmd_config(int argc, const char **argv, const char *prefix)
> {
> int nongit = !startup_info->have_repository;
> char *value;
> + int flags = 0;
>
> given_config_source.file = xstrdup_or_null(getenv(CONFIG_ENVIRONMENT));
>
> @@ -769,6 +770,8 @@ int cmd_config(int argc, const char **argv, const char *prefix)
> error(_("--fixed-value only applies with 'value_regex'"));
> usage_builtin_config();
> }
> +
> + flags = CONFIG_FLAGS_FIXED_VALUE;
I wonder whether using |= here will save someone from a headache later,
when they want to add another flag value or move the
CONFIG_FLAGS_MULTI_REPLACE calculation out of the tail calls below.
[snip]
> }
> else if (actions == ACTION_REPLACE_ALL) {
> check_write();
> @@ -850,7 +855,7 @@ int cmd_config(int argc, const char **argv, const char *prefix)
> UNLEAK(value);
> return git_config_set_multivar_in_file_gently(given_config_source.file,
> argv[0], value, argv[2],
> - CONFIG_FLAGS_MULTI_REPLACE);
> + flags | CONFIG_FLAGS_MULTI_REPLACE);
[snip]
> @@ -887,7 +893,7 @@ int cmd_config(int argc, const char **argv, const char *prefix)
> check_argc(argc, 1, 2);
> return git_config_set_multivar_in_file_gently(given_config_source.file,
> argv[0], NULL, argv[1],
> - CONFIG_FLAGS_MULTI_REPLACE);
> + flags | CONFIG_FLAGS_MULTI_REPLACE);
> --- a/config.c
> +++ b/config.c
> @@ -2402,6 +2402,7 @@ struct config_store_data {
> size_t baselen;
> char *key;
> int do_not_match;
> + const char *literal_value;
> regex_t *value_regex;
> int multi_replace;
> struct {
> @@ -2431,6 +2432,8 @@ static int matches(const char *key, const char *value,
> {
> if (strcmp(key, store->key))
> return 0; /* not ours */
> + if (store->literal_value)
> + return !strcmp(store->literal_value, value);
Nice. Short-circuiting the entire regex process if necessary.
> @@ -2803,6 +2806,8 @@ int git_config_set_multivar_in_file_gently(const char *config_filename,
> store.value_regex = NULL;
> else if (value_regex == CONFIG_REGEX_NONE)
> store.value_regex = CONFIG_REGEX_NONE;
> + else if (flags & CONFIG_FLAGS_FIXED_VALUE)
> + store.literal_value = value_regex;
Ah, so we use .literal_value instead of pulling the string from
value_regex because value_regex undergoes some special parsing and is
packed into a regex_t instead.
> else {
> if (value_regex[0] == '!') {
> store.do_not_match = 1;
> diff --git a/config.h b/config.h
> index 80844604ab..977e690be8 100644
> --- a/config.h
> +++ b/config.h
> @@ -269,6 +269,13 @@ int git_config_key_is_valid(const char *key);
> */
> #define CONFIG_FLAGS_MULTI_REPLACE (1 << 0)
>
> +/*
> + * When CONFIG_FLAGS_FIXED_VALUE is specified, match key/value pairs
> + * by string comparison (not regex match) to the provided value_regex
In contrast to my comment elsewhere, I think here it makes sense to talk
about strcmp like you did. But notice that here in the dev-facing
documentation you said "not regex" - so I think that you thought it was
a useful distinction, but forgot to say so to the user.
> + * parameter.
> + */
> +#define CONFIG_FLAGS_FIXED_VALUE (1 << 1)
> +
> diff --git a/t/t1300-config.sh b/t/t1300-config.sh
> index 6dc8117241..30e80ae9cb 100755
> --- a/t/t1300-config.sh
> +++ b/t/t1300-config.sh
> +test_expect_success '--fixed-value uses exact string matching' '
> + GLOB="a+b*c?d[e]f.g" &&
> + rm -f initial &&
This tells me that when you used 'initial' earlier you probably should
have done something like 'test_when_finished rm initial' in this test
and your earlier ones. Whoops.
> + git config --file=initial fixed.test "$GLOB" &&
> +
> + cp initial config &&
> + git config --file=config fixed.test bogus "$GLOB" &&
> + git config --file=config --list >actual &&
> + cat >expect <<-EOF &&
> + fixed.test=$GLOB
> + fixed.test=bogus
> + EOF
> + test_cmp expect actual &&
> +
> + cp initial config &&
> + git config --file=config --fixed-value fixed.test bogus "$GLOB" &&
> + git config --file=config --list >actual &&
> + printf "fixed.test=bogus\n" >expect &&
It is jarring to me to see a printf here when everywhere else we use
heredocs. 'git grep' tells me it's not unheard of, but it looks like
those are cases where the whole file doesn't use heredocs.
> + test_cmp expect actual &&
> +
> + cp initial config &&
> + test_must_fail git config --file=config --unset fixed.test "$GLOB" &&
> + git config --file=config --fixed-value --unset fixed.test "$GLOB" &&
> + test_must_fail git config --file=config fixed.test &&
Is this one supposed to verify that there is a 'fixed.test' value
already in 'config'? I'd prefer to see that explicitly checked with 'git
config --get' rather than watching for a symptom, that is, fail to set.
This comment applies to the next case too.
> +
> + cp initial config &&
> + test_must_fail git config --file=config --unset-all fixed.test "$GLOB" &&
> + git config --file=config --fixed-value --unset-all fixed.test "$GLOB" &&
> + test_must_fail git config --file=config fixed.test &&
> +
> + cp initial config &&
> + git config --file=config --replace-all fixed.test bogus "$GLOB" &&
> + git config --file=config --list >actual &&
> + cat >expect <<-EOF &&
> + fixed.test=$GLOB
> + fixed.test=bogus
> + EOF
> + test_cmp expect actual &&
Hm, isn't this the same functionality as the tests you added at the
beginning of this series? I guess you are setting up for the last case
with --replace-all...
> +
> + cp initial config &&
> + git config --file=config --replace-all fixed.test bogus "$GLOB" &&
> + git config --file=config --list >actual &&
> + cat >expect <<-EOF &&
> + fixed.test=$GLOB
> + fixed.test=bogus
> + EOF
> + test_cmp expect actual &&
Is this one identical to the previous one? I think it is, but if it
isn't and I can't tell, all the more reason that each case here should
either be labeled with a comment or separated into its own test. (Bonus
- you could extend the individual tests from patch 1 to make sure they
work correctly with --fixed-value too ;) )
> +
> + git config --file=config --fixed-value --replace-all fixed.test bogus "$GLOB" &&
> + git config --file=config --list >actual &&
> + cat >expect <<-EOF &&
> + fixed.test=bogus
> + fixed.test=bogus
Hum, it does what it says on the box, but is this a valid config, I
wonder?
$ git config --file=test foo.test aaa
$ git config --file=test --add foo.test bbb
$ git config --list --file=test
foo.test=aaa
foo.test=bbb
$ git config --file=test --replace-all foo.test bbb a+
$ git config --list --file=test
foo.test=bbb
foo.test=bbb
So I guess it's fine :)
- Emily
