[Python-Dev] Base-85

[Antoine Pitrou]

Le samedi 02 août 2008 à 14:07 -0700, Guido van Rossum a écrit :
> On Sat, Aug 2, 2008 at 12:58 PM, Antoine Pitrou <solipsis at pitrou.net> wrote:
> > It is also used by git for diffs of binary files, and those diffs are supposedly
> > understood by other VCSes like Mercurial...
> 
> I'm very interested in this (for Rietveld). Where can I learn more
> about how git handles diffs of binary files? Does it actually show
> adds and deletes of sections of the file?
> 

Well, I'm not sure. I just tried with Mercurial, first committing a
binary file with the following structure:
    part1 part3
and then changing it to the following structure:
    part1 part2 part3 part2

(part{1,2,3} being some binary chunks of 400 bytes each
from /dev/urandom)

The "git-style" diff given by Mercurial is then:

diff --git a/binfile b/binfile
index acfa6ffc5287c6e9cd400af7b8ab09d072a28b02..5b9a69212ae8f39bf41fbf2194db2b730dcb0ae9
GIT binary patch
literal 1600
zc%1Fi`#%#1003~2SQat~^D2)~Q>OB2Y-`ijkz&}wZbC#-o=0t7jaVepQ|D0_8*3gJ
zMSY0rj%^W39zC2fTEo0 at yVCs|_rrbvhcAJotX5m|hB+RX(Aa5xSa4Y^GkS%y10Hva
z3^q{I&mAF9vs at GpEP1!lCxtq*vdKD&+M&87%65P%egC&>7+Bgzx0-lUziyCW?%ELc
z;eHsAnXOY+YY~y3f6~CD+?JujZGa*JV=V-x-twhC^~z}e+->VcW=&UqfNg97Mxf3d
zP2!VM#<4|n+(B|5rOUMBfQ=w}vEdoi_TK&saEG1S{mn at ndj^rKLkR~K7EJZGGO3U9
z<g>qYkn__U%akFI(fHNxLoP?qI1sW!5 at Div?MoBQU<}W8T2DXa{`gkjO1RO?{-Yz3
z-yd-sVx%pSu0elCXI*-RPErV~&bEbl*yk6ff?mV<KJ-WwL$dK$cWOO8v8nQ6i6@~u
zvw~4tJo7xMUB<3mj-!r at xsl3A+<JySyN4MA5V)G1_~N92FPXjkyeyjjlF2Cwdfx=~
zrIy88pZDkmmlT6BVEB&HsO>pWHk^!$4x~AJ$M!$>LrjA^UDjlQD0igDHf?>kfTL8C
zT`$ToUViVhgSRSdJij81#v>3jDpj>m&hGZ(a4UQqYSc{FI)@&=mHL-D&8&McIqHsU
zfI-aCDfNmLo!AZjn=0JV{sMJcsTSiO;$}r>P(?>*s6&cpc-Lu5__+c38NPW>O=Ze3
zuFNT^b+XahK^P*sIU93zA4b<Bi<K+TWMzdB4}W%epKy!i*d>tzW?3CiviE&Xi9>Bo
zUeM*=71Le)&!sv725-*vg<uCI_{^T8FxI^)wh*sYr0H at is@*)iWzn066>S1naIlau
zItQO-*hG^8yP=|O<8TL=MWBR_ts5qYMlzBvx9 at assRs0B-+%0N-!#0o=9E#}7KwgS
zL(7(rp`ZFnuqIS(+L{IckNZ%BsAvS7nS6!Qvhd<7{m}yj9>#ZU at DGy9C0o;vRo8VN
z{7^{BvC^#ss2zI20|7B%d7&`+;}M|Lb4kcpnT&a>ztw$PNWnC?zuEIbUm<$TLPu{@
zFY}#Iu~hB^StZra*ohA%zVI_1hLduWZa6S~5X9x-A^(HQYtpfwIrX2BO0ucP;atl(
z9w7PlD`@LzhQ`qgJ$YA&U#BnZ>E7FQ)o02 at hJW4jTYGs3*i0#Kwn2EEC`iaj#%!n!
zBp#{xc{d%j`Q}GwfJ*5YN5L1jO$egiqV_Dr=2I4()xVmuAhxjtq%Tv8Xpj0eNbDC`
zr)~pTMOKjpr4)Ts!=w3K?TKMAmsB;SKU4G(tZRNi_28i83YGM!sFTf at cL9z%07l$5
zz=K?Re2pPf)sp-XNp`QKAIQ}l7Qjr5SbMa;Q&#UT4(|#}kQ>Zx58&XEdwZV}rT}gi
zAJ(u*feBc}#SFv;?WH=Nl}c*ntY4Tbs at l#;Ya(NdbFhN{QV43Y-kINHGC2 at E)ms1`
zEe~IK=-G~B2d_gQ&0mTKxO#ub8Q)9wI!6xVMVM}&lvlFHsV+BT+H~fhnnW~!EEnc+
zD=DAs=Y01@@>AfcR=Yz7$Z5mZ8P~c0<c+6^Ln)%3kuKh-Dp%rRlz3?{E)fP1KT406
Xzmcn%Pn at uyr&=uAM*jcfzxCr^k_Znj


>From that I don't know what can be done with the diff. Looking at the
Mercurial source code suggests that you can encode deltas in the patch,
but that Mercurial doesn't support it (see "# TODO: deltas"):
http://www.selenic.com/hg/index.cgi/file/cbdfd08eabc9/mercurial/patch.py#l1117

A basic explanation of binary diffs here:
http://www.selenic.com/pipermail/mercurial/2008-July/020184.html
The explanation mentions base-64 but it was corrected in a later message
here:
http://www.selenic.com/pipermail/mercurial/2008-July/020192.html

Regards

Antoine.


PS: here are the commands I've typed:

$ hg init bindiff
$ cd bindiff/
$ dd if=/dev/urandom of=part1 bs=1 count=400
[snip output]
$ dd if=/dev/urandom of=part2 bs=1 count=400
[snip output]
$ dd if=/dev/urandom of=part3 bs=1 count=400
[snip output]
$ cat part1 part3 > binfile
$ hg add binfile
$ hg ci -m "added binfile"
$ cat part1 part2 part3 > binfile
$ hg di
diff -r 19cfb10c4a01 binfile
Binary file binfile has changed
$ hg di --git
[produces the patch above]