skalibs/src, branch v2.0.0.0 The skarnet.org C programming library https://git.skarnet.org/cgit/skalibs/atom/src?h=v2.0.0.0 2014-12-22T01:04:32Z Renormalize buffer size, even if we lose 1 char 2014-12-22T01:04:32Z Laurent Bercot ska-skaware@skarnet.org 2014-12-22T01:04:32Z urn:sha1:266040fcc12715945edec022623f2dc45be31455 child_spawn1_internal fix for noposixspawn 2014-12-21T19:09:00Z Laurent Bercot ska-skaware@skarnet.org 2014-12-21T19:09:00Z urn:sha1:1fbc3336a593e7b34420172efbe81f0f4aa65ad3 OK, this one works, and the problem was actually on my side (failure to 2014-12-21T02:43:45Z Laurent Bercot ska-skaware@skarnet.org 2014-12-21T02:43:45Z urn:sha1:b8b5d27a138f32c6e400932fa8828a380a03d1d7 use CMSG_LEN where required in tryancilautoclose). Also, good news: the sys/param.h inclusion in FreeBSD isn't necessary anymore. next generation of the FreeBSD fix... 2014-12-21T02:39:19Z Laurent Bercot ska-skaware@skarnet.org 2014-12-21T02:39:19Z urn:sha1:c20608b2b2cc91367ecc2d8fe5e4cc9d4b709285 So MSG_WAITALL can block even with MSG_DONTWAIT... wtfbsdseriously. 2014-12-21T01:43:39Z Laurent Bercot ska-skaware@skarnet.org 2014-12-21T01:43:39Z urn:sha1:31009f586a04251df0fb5853017ecfcaf70b307f Only include MSG_WAITALL when it's tested nb, then. Also test yet another fix for the tryancilautoclose FreeBSD problem. Stricter trynbwaitall to actually catch the problems 2014-12-21T01:11:18Z Laurent Bercot ska-skaware@skarnet.org 2014-12-21T01:11:18Z urn:sha1:4557dacfbe04206c57e459b734016d2b6edfe7f3 So the non-portable MSG_DONTWAIT may alleviate the problem... might 2014-12-21T00:26:40Z Laurent Bercot ska-skaware@skarnet.org 2014-12-21T00:26:40Z urn:sha1:23bdcb4ad4ef19d99b729e5282fd15df6300a3b5 as well use it on architectures that support it. Something is rotten in the state of FreeBSD... but we all knew it 2014-12-20T23:15:16Z Laurent Bercot ska-skaware@skarnet.org 2014-12-20T23:15:16Z urn:sha1:828159d33b8ced9e016470c61ffd66bbef048e06 already. More tentative workarounds. FreeBSD now chokes on tryancilautoclose ; tentative fix 2014-12-20T23:02:38Z Laurent Bercot ska-skaware@skarnet.org 2014-12-20T23:02:38Z urn:sha1:7c925e3e76d239b8cca5a31ddd045aad10190e8a Found the BSD unixmessage bug. recvmsg(..., MSG_WAITALL) blocks 2014-12-20T22:25:08Z Laurent Bercot ska-skaware@skarnet.org 2014-12-20T22:25:08Z urn:sha1:db7f4d2c1ef70334af4cb8c1980f47d8a9f69249 until buffer full or socket shutdown on BSD, *even if the socket is nonblocking*. So I added an "okwaitall" sysdeps, and won't set MSG_WAITALL on retarded systems.