Re: [PATCH v3 3/9] bugreport: add version and system information

[Emily Shaffer <emilyshaffer@google.com>]

On Mon, Oct 28, 2019 at 02:49:29PM +0100, Johannes Schindelin wrote:
> Hi Emily,
> 
> On Thu, 24 Oct 2019, Emily Shaffer wrote:
> 
> > diff --git a/bugreport.c b/bugreport.c
> > new file mode 100644
> > index 0000000000..ada54fe583
> > --- /dev/null
> > +++ b/bugreport.c
> > @@ -0,0 +1,55 @@
> > +#include "cache.h"
> > +
> > +#include "bugreport.h"
> > +#include "help.h"
> > +#include "run-command.h"
> > +#include "version.h"
> > +
> > +void get_system_info(struct strbuf *sys_info)
> > +{
> > +	struct child_process cp = CHILD_PROCESS_INIT;
> > +	struct strbuf std_out = STRBUF_INIT;
> > +
> > +	strbuf_reset(sys_info);
> > +
> > +	// get git version from native cmd
> 
> Please use C-style comments instead of C++ ones.
> 
> > +	strbuf_addstr(sys_info, "git version: ");
> > +	strbuf_addstr(sys_info, git_version_string);
> > +	strbuf_complete_line(sys_info);
> > +
> > +	// system call for other version info
> > +	argv_array_push(&cp.args, "uname");
> > +	argv_array_push(&cp.args, "-a");
> > +	capture_command(&cp, &std_out, 0);
> 
> Mmmkay. I am not too much of a fan of relying on the `uname` executable,
> as it can very well be a 32-bit `uname.exe` on Windows, which obviously
> would _not_ report the architecture of the machine, but something
> misleading.
> 
> Why not use the `uname()` function (that we even define in
> `compat/mingw.c`)?

Very glad to have your review and point this kind of thing out. :) It
simplifies the code, too. I wasn't sure about these system calls, so I
appreciate you suggesting alternatives.

> 
> Also, why not include the same information as `git version
> --build-options`, most importantly `GIT_HOST_CPU`?
> 
> In any case, if you are capturing the output of `uname -a`, you
> definitely will want to pass the `RUN_COMMAND_STDOUT_TO_STDERR` flag (in
> case, say, the MSYS2 `uname.exe` fails with those dreaded Cygwin error
> messages like "*** fatal error - add_item
> ("\??\D:\git\installation\Git", "/", ...) failed, errno 1").
> 
> > +
> > +	strbuf_addstr(sys_info, "uname -a: ");
> > +	strbuf_addbuf(sys_info, &std_out);
> > +	strbuf_complete_line(sys_info);
> > +
> > +	argv_array_clear(&cp.args);
> > +	strbuf_reset(&std_out);
> > +
> > +
> > +	argv_array_push(&cp.args, "curl-config");
> > +	argv_array_push(&cp.args, "--version");
> > +	capture_command(&cp, &std_out, 0);
> 
> This will be almost certainly be incorrect, as most _regular_ users
> won't have `curl-config`. I know, it is easy to forget that most Git
> users are no hard-core C developers ;-)

Heresy! :)

> 
> A better alternative would be to use `curl_version()`, guarded, of
> course, by `#ifndef NO_CURL`...
> 
> > +
> > +	strbuf_addstr(sys_info, "curl-config --version: ");
> > +	strbuf_addbuf(sys_info, &std_out);
> > +	strbuf_complete_line(sys_info);
> > +
> > +	argv_array_clear(&cp.args);
> > +	strbuf_reset(&std_out);
> > +
> > +
> > +	argv_array_push(&cp.args, "ldd");
> > +	argv_array_push(&cp.args, "--version");
> > +	capture_command(&cp, &std_out, 0);
> 
> Again, this command will only be present in few setups. I am not
> actually sure that the output of this is interesting to begin with.

It was a suggestion, I believe, from Jonathan Nieder.

> 
> What I _do_ think is that a much more interesting piece of information
> would be the exact GLIBC version, the OS name and the OS version.

The glibc version is easy; I've done that. It certainly looks nicer than
the ldd call.

I guess I may be missing something, because as I start to look into how
to the OS info, I fall down a hole of many, many preprocessor defines to
check. If that's the approach you want me to take, just say the word,
but it will be ugly :) I suppose I had hoped the uname info would give us
a close enough idea that full OS info isn't necessary.

> 
> > +
> > +	strbuf_addstr(sys_info, "ldd --version: ");
> > +	strbuf_addbuf(sys_info, &std_out);
> > +	strbuf_complete_line(sys_info);
> > +
> > +	argv_array_clear(&cp.args);
> > +	strbuf_reset(&std_out);
> > +}
> > diff --git a/bugreport.h b/bugreport.h
> > new file mode 100644
> > index 0000000000..ba216acf3f
> > --- /dev/null
> > +++ b/bugreport.h
> > @@ -0,0 +1,7 @@
> > +#include "strbuf.h"
> > +
> > +/**
> > + * Adds the Git version, `uname -a`, and `curl-config --version` to sys_info.
> > + * The previous contents of sys_info will be discarded.
> > + */
> > +void get_system_info(struct strbuf *sys_info);
> > diff --git a/builtin/bugreport.c b/builtin/bugreport.c
> > index 2ef16440a0..7232d31be7 100644
> > --- a/builtin/bugreport.c
> > +++ b/builtin/bugreport.c
> > @@ -1,4 +1,5 @@
> >  #include "builtin.h"
> > +#include "bugreport.h"
> >  #include "stdio.h"
> >  #include "strbuf.h"
> >  #include "time.h"
> > @@ -27,6 +28,13 @@ int get_bug_template(struct strbuf *template)
> >  	return 0;
> >  }
> >
> > +void add_header(FILE *report, const char *title)
> > +{
> > +	struct strbuf buffer = STRBUF_INIT;
> > +	strbuf_addf(&buffer, "\n\n[%s]\n", title);
> > +	strbuf_write(&buffer, report);
> 
> This leaks `buffer`. Why not write into `report` via `fprintf()`
> directly?

Rather, to match the style of the rest of the builtin, modified
get_header to add the header to a passed-in strbuf instead of
modifying the file directly.

> 
> Ciao,
> Dscho
> 
> > +}
> > +
> >  int cmd_bugreport(int argc, const char **argv, const char *prefix)
> >  {
> >  	struct strbuf buffer = STRBUF_INIT;
> > @@ -43,6 +51,11 @@ int cmd_bugreport(int argc, const char **argv, const char *prefix)
> >  	get_bug_template(&buffer);
> >  	strbuf_write(&buffer, report);
> >
> > +	// add other contents
> > +	add_header(report, "System Info");
> > +	get_system_info(&buffer);
> > +	strbuf_write(&buffer, report);
> > +
> >  	fclose(report);
> >
> >  	launch_editor(report_path.buf, NULL, NULL);
> > --
> > 2.24.0.rc0.303.g954a862665-goog
> >
> >


Based on Dscho's comments, each piece of system info is gathered
differently enough that I will do each in an independent commit now.