Re: [RFC/PATCH v2 0/4] A new library for plumbing output

From: Jeff King <peff@peff.net>
Date: Apr 15, 2010, 07:15 UTC
Message-ID: <20100415071540.GB27542@coredump.intra.peff.net>
In-Reply-To: <80f140cdddc016f9b4608d79f1bc3722@212.159.54.234>

On Wed, Apr 14, 2010 at 09:57:27PM +0100, Julian Phillips wrote:

Show 8 quoted lines

> > 'git blame' has --porcelain and --incremental output, which is
> [...]
> > JSON output would only add unnecessary chatter and different quoting
> > rules.
> 
> That depends really.  If you are writing something to parse the output,
> and you already have a JSON parser available then it's the current output
> that has different quoting rules. ;)

Every once in a while, I have some crazy idea for a short script that is built around blame output (e.g., counting contributors by line count). Something that I might do in a little one-off perl script. And my experience has been that 90% of the script ends up parsing and managing commit blocks, and not the computation of interest.

Not that it's a lot of lines, mind you, but having to write 20 lines of parser to do a perl one-liner on the result is annoying. I would be very happy to have some 1 or 2 line solution where one of the lines is "use JSON;".

> Anyway, I have already converted blame to use the library for both
> --porcelain and --incremental output, so it'll be in the next version of
> the patch series.  So you can try before you buy ...

I'll be curious to see it. I hope you will (at least optionally) wrap the _whole_ output and not just the commit blocks. It would be nice to just suck it in all at once and walk the data structure. But it may be tricky because the output suppresses the commit description for commits that have already been output. You would probably want a list of lines and a map of commits, like:

  {
    "lines": [
      { sha1 and line info }
      { sha1 and line info }
      ...
    ],
    "commits": {
      "$sha1": { commit info },
      ...
  }

which is close to what I would parse to in a script, except I would actually drop the "commits" map and point directly to the commit info from each line.

Is there a way in JSON to refer to the contents of a previous item without just outputting the same data again? I assume not, and even if there is, other output formats like XML wouldn't handle it.

-Peff