On Tue, 25 Apr 2006, Dan Fandrich wrote:

> The patch has a problem in that it resets the decompression and tries
> without a header whenever the decompression fails, not just at the beginning
> of the stream. It opens the possibility (a remote one, I'll admit) of
> corrupted data getting through to the client without an error indication.

I got an added extra assignment from Paul on IRC that I added before I
committed. It makes it slightly better, although not perfect either.

As this problem seems to always occur when nothing has yet been written/
uncompressed, I figure we can probably use the writecounter to only do this at
the start of data?

> Also, there should be a comment explaining the server brokenness that this
> patch addresses.

My commit had a comment added about the reason for the weirdness.

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html