- 18 Nov, 2016 11 commits
-
-
Dridi Boukelmoune authored
-
Dridi Boukelmoune authored
-
Dridi Boukelmoune authored
-
Poul-Henning Kamp authored
-
Dridi Boukelmoune authored
-
Poul-Henning Kamp authored
CI setup.
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
Use SIGKILL which is sure to work.
-
Poul-Henning Kamp authored
-
Federico G. Schwindt authored
-
Federico G. Schwindt authored
-
- 17 Nov, 2016 1 commit
-
-
Poul-Henning Kamp authored
We recognize only the most famous signals by name SIGTERM, SIGINT and SIGKILL, if you want a special signal you will have to use "-15" or whatever. The primary reason for this is that there is still no portable API for translating "SIGXXX" to an integer. It used to be that bin/kill.c contained an array: const char *const sys_signame[NSIG] = { [0] = "Signal 0", [SIGHUP] = "HUP", [SIGINT] = "INT", [SIGQUIT] = "QUIT", [SIGILL] = "ILL", [...] BSD unix sensibly moved this into libc, to avoid duplicating this all over the place, but Linux has not done that. The right way to do this, would have been to have <signal.h> contain a table: #ifdef SIGNAL_DOC(n,s,l) SIGNAL_DOC(1, "HUP", "Hangup") SIGNAL_DOC(2, "INT", "Interrupt") ... #endif That way nobody would ever need another #ifdef SIGFOO. Anyway... Rather than pointlessly add a semi-complete list of signals no sensible person should ever use in a varnishtest (SIGWINCH anybody ?) we support the famous three by name, and the rest by number.
-
- 16 Nov, 2016 1 commit
-
-
Poul-Henning Kamp authored
-
- 15 Nov, 2016 2 commits
-
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
- 14 Nov, 2016 5 commits
-
-
Nils Goroll authored
-
Nils Goroll authored
For epoll, we tolerate spurious reports, for all other waiters we assert. fixes #2117
-
Nils Goroll authored
-
Nils Goroll authored
0 bytes are not a bogus number
-
Poul-Henning Kamp authored
-
- 11 Nov, 2016 2 commits
-
-
Nils Goroll authored
commit c457b965 was incomplete This should fix #2128 for good
-
Poul-Henning Kamp authored
-
- 09 Nov, 2016 6 commits
-
-
Nils Goroll authored
doc-fixes #2129
-
Nils Goroll authored
May be a preliminary solution should phk decide to do this differently. Fixes #2128
-
Nils Goroll authored
-
Nils Goroll authored
-
Nils Goroll authored
-
Nils Goroll authored
Related to #2130
-
- 08 Nov, 2016 4 commits
-
-
Nils Goroll authored
-
Nils Goroll authored
The otherwise recommended clock_get_time(SYSTEM_TIME) is about 40x slower than mach_absolute_time() even after minimizing overhead by retrieving the port right only once per process - see https://github.com/nigoroll/varnish-cache/tree/macos_SYSTEM_CLOCK So we revert to mach_absolute_time() which needs some scaling but otherwise does the job. It is reported to halt on iOS devices when sleeping, but even if anyone came up with the idea to sell varnish over the app store, sleeping would probably need to be avoided in the first place and otherwise the right way would probably be to adjust the offset using pre/post sleep callbacks. As MacOS should have been the last platform using the real clock fallback for VTIM_mono, we now bail out at compile-time for any platform we have overlooked.
-
Dridi Boukelmoune authored
Fixes #1899 Closes #2110
-
Dridi Boukelmoune authored
Ramp up from simple tokens to here documents.
-
- 07 Nov, 2016 2 commits
-
-
Nils Goroll authored
-
Federico G. Schwindt authored
-
- 06 Nov, 2016 3 commits
-
-
Federico G. Schwindt authored
-
Federico G. Schwindt authored
Some hardened Linux' will map the segment using random addresses as well.
-
Federico G. Schwindt authored
-
- 04 Nov, 2016 3 commits
-
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
When checking the timestamps on deleting a session, allow up to clock_step seconds error before bailing. Fixes: #1874
-
Martin Blix Grydeland authored
This value is used in calculations that could reveal clock step, specifying the amount of error we allow before panicking. It defaults to a conservative 1 second.
-