[Zlib-devel] [PATCH] One remaining Valgrind error

[Daniel Richard G.]

On Thu, 16 Aug 2012, Jan Seiffert wrote:

> As i told you already, it's the "inner loop" outside of zlib.
> Every compressed http-connection (the majority today) needs a
> new zstream (or an inflate reset). You are not just memsetting
> some 100 bytes, but prop. 64k Byte.
> So $ALEXA_WEB_TOP_10 will send a patch to remove it because their
> server handle less load/need more energy. For them it's a regression.

They're already going to be using inflateReset() in that scenario.

> And here we have the little bugger.
> But your patch was for deflateInit2_
> So call me confused.

Zlib 1.2.3.4 and 1.2.7+ give different Valgrind errors, but the latter 
still gives errors. Sorry for the mix-up---I should have posted a trace 
for current code, instead of a typical "in the wild" error that got me 
here in the first place.

> Why not? Enough libs are treated that way, esp. because these debugger 
> are not perfect.

Some libraries are handled that way, perhaps because they are buggy and 
can't be patched, or because they were not written with memory debuggers 
in mind. These debuggers, for what they do, work just fine; it's really a 
matter of interpreting the issues they detect.

> For some clients it's an performance path. If the memset would be 
> smaller, say the first 16 bytes of the array or something like that...

Maybe we could get away with initializing just a portion of s->prev, but I 
don't know how zlib works well enough to hack that.


--Daniel


-- 
Daniel Richard G. || danielg at teragram.com || Software Developer
Teragram Linguistic Technologies (a division of SAS)
http://www.teragram.com/