Re: [PATCH] run-command.c: print env vars when GIT_TRACE is set

From: Jeff King <peff@peff.net>
Date: Jan 11, 2018, 10:07 UTC
Message-ID: <20180111100734.GA17274@sigill.intra.peff.net>
In-Reply-To: <20180110104835.22905-1-pclouds@gmail.com>

On Wed, Jan 10, 2018 at 05:48:35PM +0700, Nguyễn Thái Ngọc Duy wrote:

Show 7 quoted lines

> Occasionally submodule code could execute new commands with GIT_DIR set
> to some submodule. GIT_TRACE prints just the command line which makes it
> hard to tell that it's not really executed on this repository.
> 
> Print env variables in this case. Note that the code deliberately ignore
> variables unsetting because there are so many of them (to keep git
> environment clean for the next process) and really hard to read.

I like this, and I'm pretty sure it would have helped me debug at least once in the past. I did notice one funny thing, though I think it was largely there before.

The output for a single command is pretty shell-like due to the quoting:

  $ GIT_TRACE=1 ./git upload-pack . >/dev/null
  [...]run_command: 'git-upload-pack' '.'

You could copy and paste that to a shell if you wanted. And with environment variables, that remains so:

  $ GIT_TRACE=1 ./git ls-remote https://github.com/git/git >/dev/null
  [...]run_command: 'GIT_DIR=.git' 'git-remote-https' 'https://[...]'

But if we're actually running a command via the shell, it all gets quoted as one argument:

  $ GIT_TRACE=1 GIT_PAGER='foo | bar' ./git log
  [...]run_command: 'LV=-c' 'foo | bar'

which is kind of weird, as that's not something that can be run in a shell. This isn't introduced by your patch at all, but I noticed it more because of the shell-like environment variable output.

We actually used to print a separate:

  exec: /bin/sh -c 'foo | bar'

line when we invoked a shell, which would arguably be the right place to show the env variables for such a case. But that went away with 3967e25be1 (run-command: prepare command before forking, 2017-04-19).

I think it might be helpful if we added back in "/bin/sh -c" to the run_command line when "use_shell" is in effect (and when we're not doing our "skip the shell" trickery). But that's totally orthogonal to your patch.

And anyway, it's just tracing output, so I don't think it's incredibly important either way. It was just something I noticed while looking at your patch's output.

-Peff