[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: svn corruption: trunk/packages/po/sublevel4/da.po

On Sunday 18 January 2009, Philip Hands wrote:
> Having played with the -r option, I seem to have tracked down the
> revision it's stressed about:
>   phil@cold:~/src/debian-svn/d-i$ git svn fetch -r57093:57094
>   Checksum mismatch: trunk/packages/po/sublevel4/da.po
> c9893a6364a37460e7a8ee1676d7ac0c9e85f59e expected:
> 2df48ca6c7a79d95b479c4f16eec44ba
>        got: ec0b72182b31be69dcae734015d9e316
> so it looks like the transition from 57093 to 57094 is where git
> chokes.

The cause of the git-svn failure is almost certainly the file size of 
da.po. It also looks as if this is only a problem on 32-bits systems as 
my own git-svn checkout (on an amd64 system) could handle the file 
without error.

> Frans points out that the first version of the file that looks dubious
> is 55934, and looking at this:

The root of the problem has been traced to revision r55901 where a Danish 
translation update was committed. There was a problem with the encoding 
of that file, which was partly corrected in later commits, but left 
inconsistent encoding for da in different sublevels.

This has caused the file to almost double in size with each l10n sync run:
r57090:  63485294
r57094: 126924142
r57116: 253801838

It was finally corrected in r57136 for trunk; the sublevel4/da.po file in 
the lenny branch was fixed in r57347.

Also affected are several da.po files for various individual udebs, but 
not to the same extend.

It looks like the SVN repository itself is not corrupted, but unless 
someone has some very brilliant idea how to change history it does look 
as if we'll have to live with these extremely large files until the end 
of time.

Apparently git-svn on a 32-bits system has problems with files larger than 
64 MB. This means that git-svn fetch operations for revisions between 
r57090 and r57347 can potentially fail.
I would expect checkouts that start after r57347 to work again. And 
possibly a full checkout done on a 64-bit system can be used without 
problem on a 32-bit system.

I'll also document this issue on the D-I git-svn wiki page.


Attachment: signature.asc
Description: This is a digitally signed message part.

Reply to: