Re: [Mldonkey-users] pthread MD4 hashing bug ?

[MLdonkey]

>  it seems there's a bug in MD4 hashing when it is done in a separate thread.
>  Usually I rarely see files going past the 100% downloaded mark. IIRC 
>  these files had bad chunks that had to be re-downloaded. With a pthread 
>  enabled 2.02-21, several files at once went past the 100% mark, some 
>  even hitting 120 to 130% (on 700MB sized files). "Verify chunks" did 
>  reset the downloaded amount to the real one, but these files went past 
>  the 100% again.
>  As I never had these amounts of bad chunks before, I suspected a problem 
>  in MD4 hashing in a separate thread.
>  I disabled pthread support, recompiled mldonkey and restarted with the 
>  new binary. All these downloads reached their end without me witnessing 
>  an "over 100%" downloaded percentage again.

The threads are only used to compute md4s for shared files, not for
files being downloaded. I checked the code to see if it is re-entrant,
and it is, so there is no problem if the hashes for files being
downloaded and shared files are computed concurrently. 

This problem is an old bug which has not been solved yet. We will try
to fix it before 2.03.

Thanks,

- MLDonkey