On Wed, Jun 12, 2019 at 07:29:51AM +0000, Kenth Eriksson wrote: > > Also, better way to handle 1/3 is to add condition on line io.c line > > 2211, so that if we have full buffer, we do not try to check POLLIN, > > so that we do not even call sk_read() in such case instead of try it > > to handle from inside sk_read(). > > > Ok, I can update patch 1/3 so that is not included in the pollset if > the io buffer is full. Will repost, so that you can review. But I still > don't see reading 0 shall close the socket as reading 0 is not an > error. Shall we combine the two patches to get a more robust solution?
My position is that sk_read() should depend on result of internal read() and not on external information like POLLIN status. But how to distinguish reading 0 as a result of socket close from other cases? Also, the issue of empty buffer could be simply handled by return 0 from sk_read() just before the read(). If there is no space in buffer, then there is no point to call read(). I will look at the patches and merge something reasonable. -- Elen sila lumenn' omentielvo Ondrej 'Santiago' Zajicek (email: santi...@crfreenet.org) OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net) "To err is human -- to blame it on a computer is even more so."
