On Tue, Jul 7, 2020 at 7:02 AM Jeff King [off-list ref] wrote:

On Thu, Jul 02, 2020 at 04:08:43PM +0200, Christian Couder wrote:

quoted

The other difference with V1 is that there are more tests in patch
2/2. These new tests required a small helper function to be
introduced.

I'm still not sure why %(objectsize) isn't sufficient here. Is there
some use case that's served by %(contents:size) that it wouldn't work
for? Or are we just trying to make it more discoverable when you're
looking at the contents already?

%(objectsize) is the size of the whole commit or tag object, while
%(contents:size) is the size of the complete message (the whole commit
message or tag message, including trailers and signatures).

$ git for-each-ref --format='refname: %(refname)%0a%09objectsize:
%(objectsize)%0a%09contents:size: %(contents:size)'
refs/heads/{master,next,seen} refs/tags/v2.2{5,6,7}.0
refname: refs/heads/master
        objectsize: 285
        contents:size: 69
refname: refs/heads/next
        objectsize: 281
        contents:size: 17
refname: refs/heads/seen
        objectsize: 589
        contents:size: 325
refname: refs/tags/v2.25.0
        objectsize: 974
        contents:size: 842
refname: refs/tags/v2.26.0
        objectsize: 974
        contents:size: 842
refname: refs/tags/v2.27.0
        objectsize: 974
        contents:size: 842

When sending only the complete message, not the whole object, through
a protocol, it might be interesting to easily get the length of the
complete message. For example one could use "Content-Length:
%(contents:size)%0a" when sending a complete message over HTTP.