Re: [PATCH] Add log.abbrev-commit config option

On Sat, May 14, 2011 at 3:01 PM, Jonathan Nieder [off-list ref] wrote:

Why two options?  It would be more conventional to provide just
log.abbrevCommit.  The existing "[add] ignore-errors" is explained in
the manual to be a mistake

Ah, okay.

quoted

config option as
a convenience for users who often use --abbrev-commit with git log and
git show (but not git whatchanged, as its output is more likely to be
parsed even though it is not technically plumbing).

Hm, that wouldn't have been my hunch.  Are you aware of any scripts
that parse "git whatchanged" output?

Yes, I've seen it in post-receive hooks, however, that's typically in
bare repos where I suppose it's unlikely for the user to set
log.abbrevCommit, so maybe I was just being overly cautious.

I think that's overkill. Thinking through it more, this would only
break any scripts:

a) that are parsing whatchanged output; and
b) the user sets log.abbrevCommit w/o updating his script

People using git by hand would certainly appreciate
--no-abbrev-commit, I suspect.

That's why I added it.

quoted

--- a/Documentation/config.txt
+++ b/Documentation/config.txt

@@ -1314,6 +1314,12 @@ interactive.singlekey::

      linkgit:git-checkout[1]. Note that this setting is silently
      ignored if portable keystroke input is not available.

+log.abbrev-commit::
+log.abbrevCommit::
+     If true, act as if --abbrev-commit were specified on the command
+     line. May be overridden with --no-abbrev-commit. Note that this setting
+     is ignored by rev-list.

Style: most of that page is written from the point of view of the
user

[...]

So maybe something like:

       Whether to abbreviate hexadecimal commit object names in
       output from the 'log' family of commands.  The number of
       digits shown is determined by the `--abbrev` command-line
       option and `core.abbrev` configuration variable.  Can be
       overridden on the command line by --abbrev-commit /
       --no-abbrev-commit.  The default is false.
 +
 This does not affect the 'git diff-tree' and 'git rev-list'
 commands.

I thought I phrased it like the nearby options in config.txt. and that
sounds overly verbose to me, but I'll take another look.

quoted

--- a/Documentation/git-log.txt
+++ b/Documentation/git-log.txt

@@ -38,6 +38,9 @@ OPTIONS

      Continue listing the history of a file beyond renames
      (works only for a single file).

+--no-abbrev-commit::
+     Don't abbreviate commit name. Useful for overridding log.abbrevCommit.

Also useful for overriding --abbrev-commit from aliases. :)

Shouldn't it be documented next to --abbrev-commit?

As I implemented it, --no-abbrev-commit is only honored from within
log.c, so I didn't want it to show up in the rev-list man page.

But, see below where I address your question about why I didn't
implement --no-abbrev-commit in revision.c

quoted

--- a/builtin/log.c
+++ b/builtin/log.c

@@ -23,6 +23,7 @@

 /* Set a default date-time format for git log ("log.date" config variable) */
 static const char *default_date_mode = NULL;

+static int default_abbrev_commit = 0;
 static int default_show_root = 1;
 static int decoration_style;

Style: we try to avoid unnecessary zero initializers for variables in
the BSS section.

Okay.

[...]

quoted

@@ -89,11 +91,13 @@ static void cmd_log_init_finish(int argc, const char **argv, const char *prefix,

                       struct rev_info *rev, struct setup_revision_opt *opt)
 {
      struct userformat_want w;
-     int quiet = 0, source = 0;
+     int quiet = 0, source = 0, no_abbrev_commit = 0;

      const struct option builtin_log_options[] = {
-             OPT_BOOLEAN(0, "quiet", &quiet, "supress diff output"),
+             OPT_BOOLEAN(0, "quiet", &quiet, "suppress diff output"),
              OPT_BOOLEAN(0, "source", &source, "show source"),
+             OPT_BOOLEAN(0, "no-abbrev-commit", &no_abbrev_commit,
+                         "don't abbreviate commit name"),

What happens if I do

       git log --no-abbrev-commit --abbrev-commit

?  How about

       git log --no-abbrev-commit --no-no-abbrev-commit --abbrev-commit

? :)  The behavior should be nicer if this is implemented in revision.c.

It was a thinko to put it in log.c. I added --no-abbrev-commit
thinking of its primary use case to override log.abbrevCommit. But
obviously it's more generally useful in revision.c (even though that
doesn't honor log.abbrevCommit), since it can still be used to
override earlier CLI options that might enable abbreviation.

I'll move it. Then I can document the option next to --abbrev-commit
where it makes sense.

[...]

quoted

@@ -323,6 +330,11 @@ static int git_log_config(const char *var, const char *value, void *cb)

              return git_config_string(&fmt_pretty, var, value);
      if (!strcmp(var, "format.subjectprefix"))
              return git_config_string(&fmt_patch_subject_prefix, var, value);
+     if (!strcasecmp(var, "log.abbrevcommit") ||
+         !strcasecmp(var, "log.abbrev-commit")) {

No need to use strcasecmp --- the vars passed to config functions
already have the section and variable names in lowercase.

Okay, it was a cut-and-paste from ignore-errors I think.

quoted

@@ -347,6 +359,7 @@ int cmd_whatchanged(int argc, const char **argv, const char *prefix)

      struct setup_revision_opt opt;

      git_config(git_log_config, NULL);
+     default_abbrev_commit = 0;

Puzzling as mentioned above.

Will drop.

quoted

--- a/t/t4202-log.sh
+++ b/t/t4202-log.sh

@@ -450,6 +450,14 @@ test_expect_success 'log.decorate configuration' '

 '

+test_expect_success 'log.abbrev-commit configuration' '
+     test_might_fail git config --remove-section log &&
+     git log --abbrev-commit >expect &&
+     git config log.abbrev-commit true &&
+     git log >actual &&
+     test_cmp expect actual
+'

To avoid polluting the configuration, it would be nicest to do:

       git config log.abbrev-commit true &&
       test_when_finished "git config --unset log.abbrev-commit" &&
       git log >actual &&

though it looks like some tests already protect themselves.

Other tests in t4202 protect themselves at the start, I emulated that behavior.

Just because I'm curious: what happens if you do

       git config log.abbrev-commit true

in test_create_repo in test-lib.sh?  (I.e., are there many tests that
would be confused by this?)  Tests tend to be more picky than user
scripts about the output of git but it might still be an ok way to
get a vague sense of the impact.

I'll let you know. :-)

Hope that helps, and thanks for a pleasant read.

Thanks for the quick review.

j.