From: Mark Levedahl <mlevedahl@gmail.com>
Date: Thu, 05 Jul 2007 01:58:33 GMT
Subject: Re: gitk - error in git repo on cygwin
Message-ID: <468C5049.9040403@gmail.com>
In-Reply-To: <18060.12148.313090.53861@cargo.ozlabs.ibm.com>

Paul Mackerras wrote:
> I don't think anything like that has changed.  I think it is probably
> due to the commits being batched up differently in getcommitlines.
> I'll look into it.
>
> Paul.
>
>   
Indeed, it does seem to be something order dependent. I added some debug 
printouts, and looking in proc "rowranges" when the error occurs (for 
gitk f77a29e), I get:

$commitrow($curview,$id) =  8571
[lindex $rowrangelist $commitrow($curview,$id)] = 
29504118f8528f658fd0bfc02d8d78d4c01dc2cc 8571

In all previous cases, the lindex expression evalutates to two sha1's. 
So clearly, the order of processing is different. I did check that the 
input (from git-rev-list) is identical across the linux and Cygwin 
invocations. Shrug.

Mark