Re: [PATCH v4 2/2] merge-tree.c: support --merge-base in conjunction with --stdin
- From
Elijah Newren <newren@gmail.com>
- Date
- Nov 3, 2022, 01:13 UTC
- Message-ID
- <CABPp-BG4in6hPnniHoE+au0XyVXwgs9pNUJdpLDTmP=dxvKjqw@mail.gmail.com>
- In-Reply-To
- <db47fbc663eb0f3a1fc9a063dfb1051bc64601af.1667292904.git.gitgitgadget@gmail.com>
On Tue, Nov 1, 2022 at 1:55 AM Kyle Zhao via GitGitGadget <gitgitgadget@gmail.com> wrote:
> > From: Kyle Zhao <kylezhao@tencent.com> > > The previous change add "--merge-base" option in order to allow users to
s/add/added/ ?
> specify merge-base for the merge. But it doesn't compatible well with > --stdin, because multiple batched merges can only have the same specified > base.
"it doesn't compatible well" doesn't parse for me.
> This patch allows users to pass --merge-base option into the input line, > such as:
Quoting from Documentation/SubmittingPatches:
""" Describe your changes in imperative mood, e.g. "make xyzzy do frotz" instead of "[This patch] makes xyzzy do frotz" or "[I] changed xyzzy to do frotz", as if you are giving orders to the codebase to change its behavior. """
> > printf "--merge-base=<b3> <b1> <b2>" | git merge-tree --stdin > > This does a merge of b1 and b2, and uses b3 as the merge-base.
Perhaps something like:
""" The previous commit added a `--merge-base` option in order to allow using a specified merge-base for the merge. Extend the input accepted by `--stdin` to also allow a specified merge-base with each merge requested. For example:
printf "--merge-base=<b3> <b1> <b2>" | git merge-tree --stdin
does a merge of b1 and b2, and uses b3 as the merge-base. """
However, I'm a bit curious about using `--merge-base=` on the input line as opposed to just using a simpler marker; something like
printf "<b3> -- <b1> <b2>" | git merge-tree --stdin
(which follows a precedent set by git-merge-recursive). Your version is a bit more self-documenting, but what if we want to allow users to specify multiple merge bases in the future (for use in passing to merge_incore_recursive())? Is it annoying to need to prefix each one?
Show 18 quoted lines
> Signed-off-by: Kyle Zhao <kylezhao@tencent.com> > --- > Documentation/git-merge-tree.txt | 3 ++- > builtin/merge-tree.c | 22 ++++++++++++++++++++-- > t/t4301-merge-tree-write-tree.sh | 21 +++++++++++++++++++++ > 3 files changed, 43 insertions(+), 3 deletions(-) > > diff --git a/Documentation/git-merge-tree.txt b/Documentation/git-merge-tree.txt > index d9dacb2ce54..be6a11bbaec 100644 > --- a/Documentation/git-merge-tree.txt > +++ b/Documentation/git-merge-tree.txt > @@ -66,7 +66,8 @@ OPTIONS > > --merge-base=<commit>:: > Instead of finding the merge-bases for <branch1> and <branch2>, > - specify a merge-base for the merge. > + specify a merge-base for the merge. When --stdin is passed, this > + option should be passed into the input line.
I'm not sure "passed into the input line" will be clear to readers. Perhaps we want to have --stdin documented (looks like I forgot to do that in my series; oops), mentioning the input format.
Show 12 quoted lines
> [[OUTPUT]]
> OUTPUT
> diff --git a/builtin/merge-tree.c b/builtin/merge-tree.c
> index f402b807c0f..7a8049e7b0c 100644
> --- a/builtin/merge-tree.c
> +++ b/builtin/merge-tree.c
> @@ -551,16 +551,34 @@ int cmd_merge_tree(int argc, const char **argv, const char *prefix)
>
> if (o.mode == MODE_TRIVIAL)
> die(_("--trivial-merge is incompatible with all other options"));
> + if (merge_base)
> + die(_("--merge-base should be passed into the input line"));If we change the input as noted above, the wording here may need to change too.
Show 45 quoted lines
> line_termination = '\0';
> while (strbuf_getline_lf(&buf, stdin) != EOF) {
> struct strbuf **split;
> int result;
> + const char *input_merge_base = NULL;
> + const char *arg;
>
> split = strbuf_split(&buf, ' ');
> - if (!split[0] || !split[1] || split[2])
> + if (!split[0] || !split[1])
> die(_("malformed input line: '%s'."), buf.buf);
> strbuf_rtrim(split[0]);
> - result = real_merge(&o, merge_base, split[0]->buf, split[1]->buf, prefix);
> +
> + /* parse --merge-base=<commit> option */
> + arg = split[0]->buf;
> + if (skip_prefix(arg, "--merge-base=", &arg))
> + input_merge_base = arg;
> +
> + if (input_merge_base && split[2] && !split[3]) {
> + strbuf_rtrim(split[1]);
> + result = real_merge(&o, input_merge_base, split[1]->buf, split[2]->buf, prefix);
> + } else if (!input_merge_base && !split[2]) {
> + result = real_merge(&o, NULL, split[0]->buf, split[1]->buf, prefix);
> + } else {
> + die(_("malformed input line: '%s'."), buf.buf);
> + }
> +
> if (result < 0)
> die(_("merging cannot continue; got unclean result of %d"), result);
> strbuf_list_free(split);
> diff --git a/t/t4301-merge-tree-write-tree.sh b/t/t4301-merge-tree-write-tree.sh
> index 5b0073d3dcd..aec2c00b91f 100755
> --- a/t/t4301-merge-tree-write-tree.sh
> +++ b/t/t4301-merge-tree-write-tree.sh
> @@ -860,6 +860,13 @@ test_expect_success '--stdin with both a successful and a conflicted merge' '
> test_cmp expect actual
> '
>
> +
> +test_expect_success '--merge-base is incompatible with --stdin' '
> + test_must_fail git merge-tree --merge-base=side1 --stdin 2>expect &&
> +
> + grep "^fatal: --merge-base should be passed into the input line" expect
> +'Yeah, most merge-tree options are for controlling the output, and as such they are not incompatible with --stdin. This option is, so it makes sense you need a special check for it. Looks good.
Show 21 quoted lines
> + > # specify merge-base as parent of branch2 > # git merge-tree --write-tree --merge-base=c2 c1 c3 > # Commit c1: add file1 > @@ -890,4 +897,18 @@ test_expect_success 'specify merge-base as parent of branch2' ' > ) > ' > > +test_expect_success '--stdin with both a normal merge and a merge-base specified merge' ' > + cd base-b2-p && > + printf "c1 c3\n--merge-base=c2 c1 c3" | git merge-tree --stdin >actual && > + > + printf "1\0" >expect && > + git merge-tree --write-tree -z c1 c3 >>expect && > + printf "\0" >>expect && > + > + printf "1\0" >>expect && > + git merge-tree --write-tree -z --merge-base=c2 c1 c3 >>expect && > + printf "\0" >>expect && > + test_cmp expect actual > +'
This last test seems odd. You are merely testing that the output of "git merge-tree --stdin" matches the output of repeated calls to "git merge-tree", not that the merges involved produce correct results? Maybe that's fine since earlier tests verify that individual merge-tree calls are doing the right thing, I was just a bit surprised. Maybe a comment in the code that this was your intent would be helpful?