How to "pre-populate" a git message

Once in a while, I would like to run some scripts that will automatically generate some files that I keep maintained in my repository. (Yes, I know that I shouldn't have to keep them version controlled if I can regenerate them, but it's easier to check out old versions of the files than it is to check out old versions of the repository, wait 1/2 hour to regenerate the files, and then look at them).

I also like to review the files before they get committed.

Is there any way I can "pre-populate" the commit message so that I can run my "regenerate" command, review the differences, and then commit the changes with my prewritten message?

--wpd

(+cc: Jeff and Thomas, pioneers in application of notes)

Hi!

Patrick Doyle wrote:

Not what you asked, but here's an alternative approach.

The idea is that sometimes you might not want to wait 1/2 hour for each commit and that keeping generated files in this case is more about caching than tracking content.

So, wouldn't it be nice if you could ask to remember these generated files _after_ commiting? It works like this:

An expensive command to generate some files can be run (asynchronously?) by explicit request, commit hook, or cron job. First it would check if the file is already available.

	if git notes --ref=generated/expensive.result show HEAD >/dev/null 2>&1
	then
		# already cached
		exit 0
	elif test $? != 1
	then
		# exit with error message
		exec git notes --ref=generated/expensive.result show HEAD
	fi
	# no note found; let's make one!
	... expensive operation ...
	blob=$(git hash-object -w expensive.result) &&
	git notes --ref=generated/expensive.result add -C "$blob" HEAD ||
		exit

The result could be retrieved on checkout (see the post-checkout hook) or explicitly.

	git notes --ref=generated/expensive.result show HEAD >expensive.result+
	mv expensive.result+ expensive.result

See v1.7.2-rc0~159^2~1 (diff: cache textconv output, 2010-04-01) for a similar (but not identical) example. Unfortunately the notes-cache facility used there is not currently exposed for scripted use.

Sorry for the digression.  To return to your question:

Maybe "git commit -v" or the prepare-commit-msg hook could help?

Thanks for an example. Jonathan

Am 03.12.2010 18:51 schrieb Patrick Doyle:

You can either commit the changes with the "-m <msg>" or with the "-F <file>" switch. In your case I'd suggest the latter:

1. regenerate your files and include a step that generates
    a commit message and stores that in a file, say "commit.msg".
2. review the changes
3. add all your files (except for "commit.msg", of course)
4. commit:
      $ git commit -F commit.msg

The other way would be the -m switch, but then you'd have to supply the message by yourself somehow: $ git commit -m "These files were generated at ..."

HTH,
     Dirk