Re: [PATCH] advice: refactor advise API

From: Derrick Stolee <stolee@gmail.com>
Date: Feb 10, 2020, 14:38 UTC
Message-ID: <97406f9e-b8ef-b5b9-3987-cdef843b31a5@gmail.com>
In-Reply-To: <pull.548.git.1581311049547.gitgitgadget@gmail.com>

On 2/10/2020 12:04 AM, Heba Waly via GitGitGadget wrote:

Show 8 quoted lines

> From: Heba Waly <heba.waly@gmail.com>
> 
> Add a new advise_ng function that can check the visibility of advice
> messages before printing.
>
> Currently it's very easy for the callers to miss checking the
> visibility step. Also, it makes more sense for this step to be handled
> by the advice library.

This makes the advice API much easier and its uses much cleaner. Thanks!

> Also change the advise call in tag library from advise() to advise_ng()
> to construct an example of the usage of the new API.

This is a good example case.

> +static const char turn_off_instructions[] =
> +N_("\n"
> +"Turn this message off by running\n"
> +"\"git config %s false\"");

I have mixed feelings on the use of these instructions. Perhaps at minimum the addition of these instructions could be left to a separate patch than the creation of advise_ng().

My biggest concern is that this adds unexpected noise to users who want the advice to stay. I'm calling attention to it, because this part isn't a simple refactor like the rest of the patch.

If it _does_ stay, then I recommend condensing the message to a single line. For example:

	Disable this message with "git config %d false"

Show 11 quoted lines

> +void advise_ng(const char *key, const char *advice, ...)
> +{
> +	int value = 1;
> +	struct strbuf buf = STRBUF_INIT;
> +	va_list params;
> +	const char *cp, *np;
> +	
> +	git_config_get_bool(key, &value);
> +	
> +	if(value)
> +	{

Style: spacing, and opening braces are on the same line as the if:

	if (value) {

But also, this method would be simpler if the opposite case was an early return:

	if (!value)
		return;

Then the rest could have one less indentation.

Show 16 quoted lines

> +		va_start(params, advice);
> +		strbuf_vaddf(&buf, advice, params);
> +		va_end(params);
> +
> +		strbuf_addf(&buf, turn_off_instructions, key);
> +
> +		for (cp = buf.buf; *cp; cp = np) {
> +			np = strchrnul(cp, '\n');
> +			fprintf(stderr,	_("%shint: %.*s%s\n"),
> +				advise_get_color(ADVICE_COLOR_HINT),
> +				(int)(np - cp), cp,
> +				advise_get_color(ADVICE_COLOR_RESET));
> +			if (*np)
> +				np++;
> +		}
> +		strbuf_release(&buf);

This loop looks like it was copied from advise(). Perhaps we could re-use that code better by creating a new vadvise() method that takes a va_list, and have advise() and advise_ng() call it instead? I include a patch at the end of this method that does this conversion. (Feel free to incorporate it into your next version, if you want, but be sure to add your sign-off.) Then, your advise_ng() can call these:

	vadvise(advice, params);
	advise(turn_off_instructions, key);

removing the need to re-implement the for loop.

Show 21 quoted lines

> diff --git a/t/helper/test-advise.c b/t/helper/test-advise.c
> new file mode 100644
> index 0000000000..b6ec90fd18
> --- /dev/null
> +++ b/t/helper/test-advise.c
> @@ -0,0 +1,15 @@
> +#include "test-tool.h"
> +#include "cache.h"
> +#include "advice.h"
> +
> +int cmd__advise_ng(int argc, const char **argv)
> +{
> +	if (!argv[1] || !argv[2])
> +	die("usage: %s <key> <advice>", argv[0]);
> +
> +	setup_git_directory();
> +
> +	advise_ng(argv[1], argv[2]);
> +
> +	return 0;
> +}

I definitely tend to recommend more tests than most, but perhaps this unit test is overkill? You demonstrate a good test below using a real Git command, which should be sufficient. If the "turn this message off" part gets removed, then you will still have coverage of your method. It just won't require a test change because it would not modify behavior.

Show 15 quoted lines

> diff --git a/t/t7004-tag.sh b/t/t7004-tag.sh
> index 6db92bd3ba..b7c8d41899 100755
> --- a/t/t7004-tag.sh
> +++ b/t/t7004-tag.sh
> @@ -1726,6 +1726,8 @@ test_expect_success 'recursive tagging should give advice' '
>  	hint: already a tag. If you meant to tag the object that it points to, use:
>  	hint: |
>  	hint: 	git tag -f nested annotated-v4.0^{}
> +	hint: Turn this message off by running
> +	hint: "git config advice.nestedTag false"
>  	EOF
>  	git tag -m nested nested annotated-v4.0 2>actual &&
>  	test_i18ncmp expect actual
> 
> base-commit: c7a62075917b3340f908093f63f1161c44ed1475

Thanks, -Stolee

-->8--

From: Derrick Stolee <dstolee@microsoft.com>
Date: Mon, 10 Feb 2020 09:33:20 -0500
Subject: [PATCH] advice: extract vadvise() from advise()

In preparation for a new advice method, extract a version of advise() that uses an explict 'va_list' parameter. Call it from advise() for a functionally equivalent version.

Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
---
 advice.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/advice.c b/advice.c
index 249c60dcf3..fd836332da 100644
--- a/advice.c
+++ b/advice.c
@@ -96,15 +96,12 @@ static struct {
 	{ "pushNonFastForward", &advice_push_update_rejected }
 };
 
-void advise(const char *advice, ...)
+static void vadvise(const char *advice, va_list params)
 {
 	struct strbuf buf = STRBUF_INIT;
-	va_list params;
 	const char *cp, *np;
 
-	va_start(params, advice);
 	strbuf_vaddf(&buf, advice, params);
-	va_end(params);
 
 	for (cp = buf.buf; *cp; cp = np) {
 		np = strchrnul(cp, '\n');
@@ -118,6 +115,14 @@ void advise(const char *advice, ...)
 	strbuf_release(&buf);
 }
 
+void advise(const char *advice, ...)
+{
+	va_list params;
+	va_start(params, advice);
+	vadvise(advice, params);
+	va_end(params);
+}
+
 int git_default_advice_config(const char *var, const char *value)
 {
 	const char *k, *slot_name;

-- 
2.25.0.vfs.1.1.1.g9906319d24.dirty