Re: [PATCH] revert: add --stdin option to read commits from stdin

On Monday 14 June 2010 07:20:27 Jonathan Nieder wrote:

Christian Couder wrote:

quoted

--- a/t/t3508-cherry-pick-many-commits.sh
+++ b/t/t3508-cherry-pick-many-commits.sh

@@ -92,4 +92,14 @@ test_expect_failure 'cherry-pick -3 fourth works' '
 	test "$(git rev-parse --verify HEAD)" != "$(git rev-parse --verify

fourth)" '

+test_expect_success 'cherry-pick --stdin works' '
+	git checkout master &&
+	git reset --hard first &&

[...]

This test fails for me as written, since the previous test leaves some
files in an unmerged state.  Patch 1 below works around that.

Ok.

quoted

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

@@ -79,7 +80,7 @@ static void parse_args(int argc, const char **argv)
 	}

 	commit_argc = parse_options(argc, argv, NULL, options, usage_str, 0);

[...]

quoted

@@ -527,10 +528,12 @@ static void prepare_revs(struct rev_info *revs)
 {
 	int argc = 0;
 	int i;
-	const char **argv = xmalloc((commit_argc + 4) * sizeof(*argv));
+	const char **argv = xmalloc((commit_argc + 5) * sizeof(*argv));

 	argv[argc++] = NULL;
 	argv[argc++] = "--no-walk";
+	if (read_stdin)
+		argv[argc++] = "--stdin";

Ah, I see the problem now.  But it would be even nicer to allow arbitrary
rev-list options, so a person could ‘git cherry-pick --reverse a..b’,
for example.

Yeah, I agree that is nicer.

In other words, how about something like patch 2 below?

Patch 3 is a small cleanup, as a bonus.

Christian Couder (1):
  revert: accept arbitrary rev-list options

Jonathan Nieder (2):
  t3508 (cherry-pick): futureproof against unmerged files
  revert: do not rebuild argv on heap

I get "indent with spaces" errors when I apply patches 2/3 and 3/3:

$ git am ../messages/\[PATCH\ 2_3\]\ revert_\ accept\ arbitrary\ rev-list\ 
options.mbox
Applying: revert: accept arbitrary rev-list options                                                                      
/home/christian/work/git/.git/rebase-apply/patch:35: indent with spaces.                                                 
                                    PARSE_OPT_KEEP_UNKNOWN);  

$ git am ../messages/\[PATCH\ 3_3\]\ revert_\ do\ not\ rebuild\ argv\ on\ 
heap.mbox
Applying: revert: do not rebuild argv on heap                                                                        
/home/christian/work/git/.git/rebase-apply/patch:33: indent with spaces.                                             
                                    PARSE_OPT_KEEP_ARGV0 |                                                           
warning: 1 line adds whitespace errors.

But otherwise it all looks very good to me.

Even something like "git cherry-pick -3 fourth" is now working after patch 
2/3, so all test cases now pass. This is because "--no-walk" does not always 
take over "-3" it looks like it depends on the order of the arguments.

For example I get:

$ git rev-list --no-walk -3 --reverse fourth
453a04748224b3f212580d1195b452334d346e75
e85abe28a2b8ef771f760575b325f4c41f9c815f
94d3184b3f0dcfebb393faf5a122dc429d775538

but:

$ git rev-list -3 --no-walk --reverse fourth
94d3184b3f0dcfebb393faf5a122dc429d775538

I will post an updated v2 series without the whitespace errors and with a few 
documentation and test updates in patch 2/3.

Thanks,
Christian.