Large patches in git repositories

[Paweł Sikora]

On Friday 05 of October 2012 16:05:16 Kacper Kornet wrote:
> Recently one email to pld-cvs was blocked due to large size of generated
> diff. The "offending" file was gcc-branch.diff from crossmingw64-gcc
> which is 11M.
> 
> Does this file include and PLD specific changes or is it just a diff
> between released version and tip of branch in some foreign (not PLD)
> repository? In latter case maybe it should be kept in the distfiles in
> compressed form?
> 
> 
> Right now there are following limits file sizes in git
> repositories:
> 
> 1) unlimited - files *.patch, *.diff, *.spec
> 2) 2000000 bytes  - text files
> 3) 200000 bytes - all other files
> 
> so very big patches are accepted. However I wonder if it should be the
> case.
> 
> Another question is if we decide to move it to distfiles should the git
> repo be rewritten to not include history of this file and reduce repo
> size.
> 
> There are other similar files. Below I include the list of every file
> larger then 1M which its maximum size in history:
> 
> 24145161 gcc.git/gcc-branch.diff
> 22482085 crossmingw64-gcc.git/gcc-branch.diff

from branch.diff/pr-XXXX.patch point of view, ideally would be to have
an improved %patch macro called %something with support for mirrored
source svn/hg/git repositories (similary to distfiles).

e.g. in gcc.spec:

RepositoryX: svn://gcc...
%prep
// unpack released tarball
%somethingX revision-range-to-apply

such construciton should generate patch on the fly (basing on origin repo or mirrored one)
and apply it. there's no sense in rolling big stable/branch patches through versioned plain format.

one problem i can see is the .src.rpm (generated patch vs. archived .spec).