Re: [PATCH v2 08/10] builtin/bugreport.c: create '--diagnose' option

From: Victoria Dye <vdye@github.com>
Date: Aug 9, 2022, 23:53 UTC
Message-ID: <a45f5693-7186-2953-6620-3f1359a12238@github.com>
In-Reply-To: <3dc402e1-1f27-8a24-544d-d90d403a7da0@github.com>

Derrick Stolee wrote:

Show 17 quoted lines

> On 8/3/2022 9:45 PM, Victoria Dye via GitGitGadget wrote:
>> From: Victoria Dye <vdye@github.com>
> 
>> +--diagnose[=(basic|all)]::
>> +	Create a zip archive of information about the repository including logs
> 
> logs? I think the reflogs are not included unless "all" is specified. Perhaps
> we can unify this description with the beginning of git-diagnose.txt:
> 
>   Collects detailed information about the user's machine, Git client, and
>   repository state and packages that information into a zip archive.
> 
> resulting in
> 
> 	Create a zip archive containing information about the user's machine,
> 	Git client, and repository state.
>

[...]

Show 15 quoted lines

>> +	and certain statistics describing the data shape of the repository. The
>> +	archive is written to the same output directory as the bug report and is
>> +	named 'git-diagnostics-<formatted suffix>'.
>> ++
>> +By default, `--diagnose` (equivalent to `--diagnose=basic`) will collect only
>> +statistics and summarized data about the repository and filesystem. Specifying
>> +`--diagnose=all` will create an archive with the same contents generated by `git
>> +diagnose --all`; this archive will be much larger, and will contain potentially
>> +sensitive information about the repository. See linkgit:git-diagnose[1] for more
>> +details on the contents of the diagnostic archive.
> 
> Perhaps here (and git-diagnose.txt) should be really explicit about sharing the
> "all" mode output only with trusted parties. Let the user decide what level of
> trust is necessary depending on their situation (we don't need to say "open source
> repos are fine to share" or something).

Both of these documentation suggestions make sense to me; I'll update accordingly in V3.

Show 15 quoted lines

> 
>> +enum diagnose_mode {
>> +	DIAGNOSE_NONE,
>> +	DIAGNOSE_BASIC,
>> +	DIAGNOSE_ALL
>> +};
> 
> This enum makes me think that it might be nice to use this in diagnose.h
> along with an array that pairs strings with the enum. We could unify the
> options by having 'git diagnose --mode=(basic|all)' which could be
> extended in the future with another mode that might be in between the two.
> 
> It may also be a waste of time to set up that infrastructure without it
> actually mattering in the future, but I thought I'd mention it as an
> alternative, in case that inspires you.

Your suggestion is more extensible than the boolean "include_everything" I'm using right now in 'create_diagnostics_archive()'. I'll incorporate it into my next re-roll, thanks!

Show 20 quoted lines

> 
>>  static void get_system_info(struct strbuf *sys_info)
>>  {
>> @@ -91,6 +97,23 @@ static void get_header(struct strbuf *buf, const char *title)
>>  	strbuf_addf(buf, "\n\n[%s]\n", title);
>>  }
>>  
>> +static int option_parse_diagnose(const struct option *opt,
>> +				 const char *arg, int unset)
>> +{
>> +	enum diagnose_mode *diagnose = opt->value;
>> +
>> +	BUG_ON_OPT_NEG(unset);
>> +
>> +	if (!arg || !strcmp(arg, "basic"))
>> +		*diagnose = DIAGNOSE_BASIC;
>> +	else if (!strcmp(arg, "all"))
>> +		*diagnose = DIAGNOSE_ALL;
> 
> Should we allow "none" to reset the value to DIAGNOSE_NONE?

As far as I can tell, while some builtins have options that match the default behavior of the command (e.g., '--no-autosquash' in 'git rebase'), those options typically exist to override a config setting (e.g., 'rebase.autosquash'). No config exists for 'bugreport --diagnose' (and I don't think it would make sense to add one), so '--diagnose=none' would only be used to override another '--diagnose' specification in the same command/alias (e.g., 'git bugreport --diagnose=basic --diagnose=none').

That use case seems unlikely, but if there's precedent or use cases I'm not accounting for, I'm happy to add the option.

Show 7 quoted lines

> 
>> +	else
>> +		die(_("diagnose mode must be either 'basic' or 'all'"));
> 
> I wondered initially if this should be a usage() call instead. But we have
> plenty of examples of using die() to report an issue with a single option
> or a combination of options.

After looking at other 'OPT_CALLBACK_F' examples (e.g., the 'diff_opt_stat' options in 'diff.c'), I think this should at least change to 'return error(<something>)' to get a more appropriate return code (129):

	else
		return error(_("invalid --%s mode '%s'"), opt->long_name, arg);

I also looked into using 'usage_msg_opt()' to print both an error message and the usage string, but doing so would require making the 'bugreport' options structure & variables static so they're accessible by 'option_parse_diagnose()'. If you think it would be valuable to include that information, I'll add an extra refactoring patch to include it.

Show 18 quoted lines

> 
>>  	const struct option bugreport_options[] = {
>> +		OPT_CALLBACK_F(0, "diagnose", &diagnose, N_("(basic|all)"),
>> +			       N_("create an additional zip archive of detailed diagnostics"),
>> +			       PARSE_OPT_NONEG | PARSE_OPT_OPTARG, option_parse_diagnose),
> 
> The biggest reason for this to be an OPT_CALLBACK_F is because of the
> '--diagnose' option (without '='), so an OPT_STRING would not be
> appropriate here.
> 
>> @@ -119,6 +147,7 @@ int cmd_bugreport(int argc, const char **argv, const char *prefix)
>>  					    option_output ? option_output : "");
>>  	strbuf_addstr(&report_path, prefixed_filename);
>>  	strbuf_complete(&report_path, '/');
>> +	output_path_len = report_path.len;
> 
> Perhaps this should be renamed to output_dir_len, since we know this is
> a directory that will contain all of the output files.

Agreed, 'path' is a bit ambiguous and 'dir' more clearly refers to the directory containing the output files.

Show 16 quoted lines

> 
>> +	/* Prepare diagnostics, if requested */
>> +	if (diagnose != DIAGNOSE_NONE) {
>> +		struct strbuf zip_path = STRBUF_INIT;
>> +		strbuf_add(&zip_path, report_path.buf, output_path_len);
>> +		strbuf_addstr(&zip_path, "git-diagnostics-");
>> +		strbuf_addftime(&zip_path, option_suffix, localtime_r(&now, &tm), 0, 0);
>> +		strbuf_addstr(&zip_path, ".zip");
>> +
>> +		if (create_diagnostics_archive(&zip_path, diagnose == DIAGNOSE_ALL))
> 
> (Just pausing to say this could be create_diagnostics_archive(&zip_path, diagnose)
> if we use the enum inside diagnose.c.
> 
> Thanks,
> -Stolee