[PATCH] merge: use skip_prefix to parse config key

[PATCH] merge: use skip_prefix to parse config key

[Martin Ågren]

Instead of using `starts_with()`, the magic number 7, `strlen()` and a
fair number of additions to verify the three parts of the config key
"branch.<branch>.mergeoptions", use `skip_prefix()` to jump through them
more explicitly.

We need to introduce a new variable for this (we certainly can't modify
`k` just because we see "branch."!). With `skip_prefix()` we often use
quite bland names like `p` or `str`. Let's do the same. If and when this
function needs to do more prefix-skipping, we'll have a generic variable
ready for this.

Adjust the indentation while we're here.

Signed-off-by: Martin Ågren <martin.agren@gmail.com>
---

 worktree.c also matches "strcmp.*strlen" and could see a similar patch.
 I'm working on a longer series for that file and this fell out of that
 work. This feels independent enough that I'm posting it on its own.

 builtin/merge.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/builtin/merge.c b/builtin/merge.c
index d127d2225f..bde5f14f05 100644
--- a/builtin/merge.c
+++ b/builtin/merge.c
@@ -597,10 +597,10 @@ static void parse_branch_merge_options(char *bmo)
 static int git_merge_config(const char *k, const char *v, void *cb)
 {
 	int status;
+	const char *str;
 
-	if (branch && starts_with(k, "branch.") &&
-		starts_with(k + 7, branch) &&
-		!strcmp(k + 7 + strlen(branch), ".mergeoptions")) {
+	if (branch && skip_prefix(k, "branch.", &str) &&
+	    skip_prefix(str, branch, &str) && !strcmp(str, ".mergeoptions")) {
 		free(branch_mergeoptions);
 		branch_mergeoptions = xstrdup(v);
 		return 0;
-- 
2.26.0

Re: [PATCH] merge: use skip_prefix to parse config key

[Jeff King]

On Fri, Apr 10, 2020 at 05:10:32PM +0200, Martin Ågren wrote:

> Instead of using `starts_with()`, the magic number 7, `strlen()` and a
> fair number of additions to verify the three parts of the config key
> "branch.<branch>.mergeoptions", use `skip_prefix()` to jump through them
> more explicitly.

The conversion looks correct to me and is certainly an improvement.

> We need to introduce a new variable for this (we certainly can't modify
> `k` just because we see "branch."!). With `skip_prefix()` we often use
> quite bland names like `p` or `str`. Let's do the same. If and when this
> function needs to do more prefix-skipping, we'll have a generic variable
> ready for this.

I was about to comment on this in the patch, but your explanation
preempted me. :) The logic here makes sense.

> diff --git a/builtin/merge.c b/builtin/merge.c
> index d127d2225f..bde5f14f05 100644
> --- a/builtin/merge.c
> +++ b/builtin/merge.c
> @@ -597,10 +597,10 @@ static void parse_branch_merge_options(char *bmo)
>  static int git_merge_config(const char *k, const char *v, void *cb)
>  {
>  	int status;
> +	const char *str;
>  
> -	if (branch && starts_with(k, "branch.") &&
> -		starts_with(k + 7, branch) &&
> -		!strcmp(k + 7 + strlen(branch), ".mergeoptions")) {
> +	if (branch && skip_prefix(k, "branch.", &str) &&
> +	    skip_prefix(str, branch, &str) && !strcmp(str, ".mergeoptions")) {
>  		free(branch_mergeoptions);
>  		branch_mergeoptions = xstrdup(v);
>  		return 0;

In general, parsing subsections accurately involves looking from both
the start and the end of the string, pulling out the section and key and
leaving the rest in the middle. But I think we can get away with this
left-to-right parsing because we're only interested in matching a
_specific_ subsection name, and a specific key. So there are no cases it
will handle incorrectly.

The more general form would be:

  const char *subsection, *key;
  int subsection_len;

  if (!parse_config_key("branch", &subsection, &subsection_len, &key) &&
      subsection_len == strlen(branch) && !strncmp(subsection, branch) &&
      !strcmp(key, "mergeoptions"))
         ...

but that's a bit more awkward (it would be less so if we had a helper
function for comparing a NUL-terminated string against a ptr/len pair).

-Peff

Re: [PATCH] merge: use skip_prefix to parse config key

[Junio C Hamano]

Jeff King <peff@peff.net> writes:

> In general, parsing subsections accurately involves looking from both
> the start and the end of the string, pulling out the section and key and
> leaving the rest in the middle. But I think we can get away with this
> left-to-right parsing because we're only interested in matching a
> _specific_ subsection name, and a specific key. So there are no cases it
> will handle incorrectly.

In other words, if k were "branch.A.B.mergeoptions", it can only be
the 'branch.*.mergeoptions' variable attached to branch "A.B", but
when checking for branch=="A", the first two skip_prefix() would
pass and the only thing that protects us from misparsing is that
"B.mergeoptions" is not what we are looking for.

> The more general form would be:
>
>   const char *subsection, *key;
>   int subsection_len;
>
>   if (!parse_config_key("branch", &subsection, &subsection_len, &key) &&
>       subsection_len == strlen(branch) && !strncmp(subsection, branch) &&
>       !strcmp(key, "mergeoptions"))
>          ...
>
> but that's a bit more awkward (it would be less so if we had a helper
> function for comparing a NUL-terminated string against a ptr/len pair).

Yes, but even with such a helper, i.e.

	if (branch &&
	    !parse_config_key("branch", &sub, &sublen, &key) &&
	    !spanstrcmp(sub, sublen, branch) &&
	    !strcmp(key, "mergeoptions"))

what Martin wrote, especially if it is reflowed to match the above, i.e.

	if (branch &&
	    skip_prefix(key, "branch.", &sub) &&
	    skip_prefix(sub, branch, &sub) &&
	    !strcmp(sub, ".mergeoptions")

I find it just as, if not more, easy to read.

Where the parse_config_key() helper shines, I think, is when we do
not have the middle level to compare against, and in that case, we
must work only from the given key, scanning from both ends for dot.

Thanks.

Re: [PATCH] merge: use skip_prefix to parse config key

[Jeff King]

On Fri, Apr 10, 2020 at 09:44:56AM -0700, Junio C Hamano wrote:

> Jeff King <peff@peff.net> writes:
> 
> > In general, parsing subsections accurately involves looking from both
> > the start and the end of the string, pulling out the section and key and
> > leaving the rest in the middle. But I think we can get away with this
> > left-to-right parsing because we're only interested in matching a
> > _specific_ subsection name, and a specific key. So there are no cases it
> > will handle incorrectly.
> 
> In other words, if k were "branch.A.B.mergeoptions", it can only be
> the 'branch.*.mergeoptions' variable attached to branch "A.B", but
> when checking for branch=="A", the first two skip_prefix() would
> pass and the only thing that protects us from misparsing is that
> "B.mergeoptions" is not what we are looking for.

Yes, exactly.

> > The more general form would be:
> [...]
> Yes, but even with such a helper, i.e.
> [...]
> what Martin wrote, especially if it is reflowed to match the above, i.e.
> [...]
> I find it just as, if not more, easy to read.

Yeah, sorry if I was unclear; I think the left-to-right is perfectly
fine for this case.

> Where the parse_config_key() helper shines, I think, is when we do
> not have the middle level to compare against, and in that case, we
> must work only from the given key, scanning from both ends for dot.

Agreed.

Another option for known-value cases like this is to do it outside of
the callback handler, like:

  char *key = xstrfmt("branch.%s.mergeoptions");
  const char *value;
  if (!git_config_get_string_const(key, &value))
     ...
  free(key);

The allocation is a bit awkward, though we could hide that with a clever
helper.

Shifting from "iterate over and store config keys" to "look up config
keys on demand" is a much bigger change, though. I would only do it if
it made the rest of the code flow easier.

-Peff

Re: [PATCH] merge: use skip_prefix to parse config key

[Junio C Hamano]

Jeff King <peff@peff.net> writes:

> Shifting from "iterate over and store config keys" to "look up config
> keys on demand" is a much bigger change, though. I would only do it if
> it made the rest of the code flow easier.

Yeah, absolutely.

Thanks.

[PATCH v2] merge: use skip_prefix to parse config key

[Martin Ågren]

On Fri, 10 Apr 2020 at 18:56, Jeff King <peff@peff.net> wrote:
>
> On Fri, Apr 10, 2020 at 09:44:56AM -0700, Junio C Hamano wrote:
>
> > Jeff King <peff@peff.net> writes:
> >
> > In other words, if k were "branch.A.B.mergeoptions", it can only be
> > the 'branch.*.mergeoptions' variable attached to branch "A.B", but
> > when checking for branch=="A", the first two skip_prefix() would
> > pass and the only thing that protects us from misparsing is that
> > "B.mergeoptions" is not what we are looking for.
>
> Yes, exactly.
>
> > > The more general form would be:
> > [...]
> > Yes, but even with such a helper, i.e.
> > [...]
> > what Martin wrote, especially if it is reflowed to match the above, i.e.
> > [...]
> > I find it just as, if not more, easy to read.
>
> Yeah, sorry if I was unclear; I think the left-to-right is perfectly
> fine for this case.

Thanks both for an interesting discussion. Junio left a remark about
"[easy to read] especially if it is reflowed" which I recognized from my
own thinking yesterday. Then I went for fewer lines in what arguably
looks a bit crammed. Since I was not the only one to find readability in
the larger number of lines, here it is as v2.

Junio, if you go "meh, what I've already picked up will do just as
fine", feel free to act on that, i.e., to not act on this v2.

> Another option for known-value cases like this is to do it outside of
> the callback handler, like:
>
>   char *key = xstrfmt("branch.%s.mergeoptions");
>   const char *value;
>   if (!git_config_get_string_const(key, &value))
>      ...
>   free(key);
>
> The allocation is a bit awkward [...]

I very briefly thought about such an approach here, but still in the
callback handler, where the allocation would be quite a bit more
awkward, since we'd do it over and over.

Something I considered a bit more seriously was a helper that takes
three strings ("branch", branch and "mergeoptions") and does the check.
But as you mention in this thread, it's a bit special that we're looking
for a specific subsection (that is not a compile-time constant). So it
felt like the kind of "nice-to-have" helper that you'd end up using just
once or twice. (That said, I didn't look to see if I could find other
instances.)

Martin

-- >8 --
Instead of using `starts_with()`, the magic number 7, `strlen()` and a
fair number of additions to verify the three parts of the config key
"branch.<branch>.mergeoptions", use `skip_prefix()` to jump through them
more explicitly.

We need to introduce a new variable for this (we certainly can't modify
`k` just because we see "branch."!). With `skip_prefix()` we often use
quite bland names like `p` or `str`. Let's do the same. If and when this
function needs to do more prefix-skipping, we'll have a generic variable
ready for this.

Signed-off-by: Martin Ågren <martin.agren@gmail.com>
---
 builtin/merge.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/builtin/merge.c b/builtin/merge.c
index d127d2225f..df83ba2a80 100644
--- a/builtin/merge.c
+++ b/builtin/merge.c
@@ -597,10 +597,12 @@ static void parse_branch_merge_options(char *bmo)
 static int git_merge_config(const char *k, const char *v, void *cb)
 {
 	int status;
+	const char *str;
 
-	if (branch && starts_with(k, "branch.") &&
-		starts_with(k + 7, branch) &&
-		!strcmp(k + 7 + strlen(branch), ".mergeoptions")) {
+	if (branch &&
+	    skip_prefix(k, "branch.", &str) &&
+	    skip_prefix(str, branch, &str) &&
+	    !strcmp(str, ".mergeoptions")) {
 		free(branch_mergeoptions);
 		branch_mergeoptions = xstrdup(v);
 		return 0;
-- 
2.26.0