Hi! On Sat, 2009-10-24 at 17:45:03 -0500, Jonathan Nieder wrote: > The current callers for the compression code do not install signal > handlers, so there is no occasion to test the EINTR handling. > Perhaps for this reason, since commit 7bf6e0 (add support for using > libz, 2000-12-09) when the current compression/decompression code > was introduced, the EINTR handling has been broken in a number of > ways: > > * interrupted reads were treated as end of file until very > recently > * interrupted writes during decompression cause portions of the > output to be discarded > * interrupted writes during compression are treated as errors, > unless the interruption happens before any data from the > output buffer can be consumed > > Since zlib at least cannot recover from an interrupted write > anyway, it seems better to always treat EINTR like any other error. > Callers should specify the SA_RESTART flag when installing signal > handlers for correct behavior on System V style operating systems > (such as Solaris).
I've applied this one, as it's best to explicitly fail, than to try to handle a case that would cause subtle errors, even if the code currently should not be susceptible to this problem. As I said on another mail though, I might switch from using the libraries IO functions to use the IOless compression code combined with our buffer API. thanks, guillem -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
