I think it is ok for developing with git developers. But we should include the binaries in released zip file.

How to do it:

Search for "remove binary files from git history" in google you will find:

1)
https://gist.github.com/shennyg/1047737

2)
http://stackoverflow.com/questions/17418514/remove-all-binary-files-recursively-from-git-repo-and-commit-history

I finally compressed .git folder from 730.8 MB to 285.8 MB
The way is:

java -jar ~/Software/bfg.jar -b 1M -D '.{obj,o,exe,dll,apk,so,a,ipa,pdb,ilk,exp,lib,suo,ncb,sdf,sbr,opensdf,swp,swo,png,jpg,jpeg,tiff,mp3,mp4,ogg,wav,caf,pvr,zip,rar,ccz,gz,webp,ttf,bmp}'
git reflog expire --expire=now --all
git gc --prune=now --aggressive

Did this data fix our needs?

But the SHA id will be changed for every commits.

285 is much better than 730. It is 2.5 times smaller.

For sure, that is a step in the right direction.
but would it be possible to make it under 100mb ?

Thanks!

Should we run that command in all the branches ? or that command automatically performs that operation on all the branches ?

thanks!

I only checked out a v3 branch in my clean repo for testing.
I will test again for v1, v2 and v3.

ok, thanks. yes, we need to remove the huge binaries from all all the branches.

but would it be possible to make it under 100mb ?

Maybe it could be achieved by cutting some oldest commits off.

ok, thanks. yes, we need to remove the huge binaries from all all the branches.

I used that tool to remove file which is larger than 1M and which has extension of exe,dll,apk,so,a,ipa,pdb,ilk,exp,lib,suo,ncb,sdf,sbr,opensdf,swp,swo,png,jpg,jpeg,tiff,mp3,mp4,ogg,wav,caf,pvr,zip,rar,ccz,gz,webp,ttf,bmp

I wouldn't remove old commits.
In any case I would remove the "v1" branch. Nobody is using it.

Updated:

Finally, I made the repo size reduced to 146M. It seems that there are no more gaps to make it smaller.

You could clone my repo to test it.

git clone https://github.com/angeltown/cocos2d-x.git
Cloning into 'cocos2d-x'...
remote: Counting objects: 203896, done.
remote: Compressing objects: 100% (65972/65972), done.
remote: Total 203896 (delta 135728), reused 195255 (delta 130873)
Receiving objects: 100% (203896/203896), 146.59 MiB | 672.00 KiB/s, done.
Resolving deltas: 100% (135728/135728), done.
Checking connectivity... done.

The command i used is:

java -jar ~/Software/bfg.jar -b 200K -D '*.{plist,json,ExportJson,z,obj,o,exe,dll,apk,so,a,ipa,pdb,ilk,exp,lib,suo,ncb,sdf,sbr,opensdf,swp,swo,atlas,bmp,caf,ccb,ccbi,ccbproj,ccbresourcelog,ccz,dds,fnt,fsh,gz,jpeg,jpg,ktx,mid,mp3,ogg,pkm,png,ppng,pvr,pvrraw,raw,tga,tiff,tmx,tps,tsx,ttf,wav,webp,zss,ztp}'
git reflog expire --expire=now --all
git gc --prune=now --aggressive

Maybe it could be better. I will try another way.

thanks. not bad :)
but it is still a bit slow for me

I'm wondering if the test assets could be improved.
As an example, I know that we moved resources from directory to another many times.
The move operations (remove + add) were done in one single commit ? or in two (remove first, and then add) ?

If those operations were done one commit, then git will will treat it as a "rename"... otherwise, it will tread it as a "remove", and then "add".... and that will make the git history pretty big.

yes, i think it should work too.
I'm trying to strip the history for cpp-test resources

Eventually, repo size is about 50M now.
You could use https://github.com/angeltown/cocos2d-x.git to test.
But I will re-add resources which was removed. The size may be increased to about 80M.

git clone git://github.com/angeltown/cocos2d-x.git
Cloning into 'cocos2d-x'...
remote: Counting objects: 197081, done.
remote: Compressing objects: 100% (57377/57377), done.
remote: Total 197081 (delta 132402), reused 184353 (delta 131784)
Receiving objects: 100% (197081/197081), 56.79 MiB | 614.00 KiB/s, done.
Resolving deltas: 100% (132402/132402), done.
Checking connectivity... done.

Ok. thanks. well done!