Commit graph

3323 commits

Author SHA1 Message Date
Juan Jose Garcia Ripoll
5ae1c96929 The gc library does not remember its settings: we have to define GC_PTHREADS or GC_WIN32_THREADS before including the gc.h header. 2009-10-01 12:01:26 +02:00
Juan Jose Garcia Ripoll
5618cc648b src/c/threads.d: we cannot create locks until the environment is set up 2009-10-01 11:36:34 +02:00
Juan Jose Garcia Ripoll
2bee22b015 Note the last change in the configuration process, related to the detection of GC_register_my_thread. 2009-10-01 10:27:31 +02:00
Juan Jose Garcia Ripoll
5d2002432a The test for an existing Boehm-Weiser garbage collector must include function GC_register_my_thread 2009-10-01 10:20:21 +02:00
Juan Jose Garcia Ripoll
3913c4464e Functions returning "void" can not use "return something" in Sun's compiler, even if that something is (void). 2009-09-29 21:31:02 +02:00
Juan Jose Garcia Ripoll
d498548b07 src/c/package.d: field hash.lock is not defined in single-threaded versions 2009-09-29 17:04:34 +02:00
Juan Jose Garcia Ripoll
0e7563952e Change needed for OS X 10.6 Leopard: ucontext.h is deprecated and produces an error. 2009-09-29 17:04:11 +02:00
Juan Jose Garcia Ripoll
04c7b4e67b threads_mutex.d was not added in the previous commit 2009-09-27 21:52:04 +02:00
Juan Jose Garcia Ripoll
d3ce2ef931 Split the code for mutexes into a separate file and use these lock objects in cl_core, instead of the raw versions. 2009-09-27 21:20:01 +02:00
Juan Jose Garcia Ripoll
037cc67c48 Sometimes file_listen() returns that there exist characters, but actually there is an EOF. 2009-09-27 21:19:16 +02:00
Juan Jose Garcia Ripoll
500f324754 Following SBCL, use a single global lock for packages, instead of one lock per package. 2009-09-27 13:19:30 +02:00
Juan Jose Garcia Ripoll
dc6d0659b2 ecl_extend_hashtable() now does not destructively modify a package hash table. Instead, it returns a new object. This affects ecl_sethash() and all package functions that modify the package tables. 2009-09-27 10:46:54 +02:00
Juan Jose Garcia Ripoll
ae9515ee2b Reorganize the list of latest changes. 2009-09-27 10:44:25 +02:00
Juan Jose Garcia Ripoll
95f98db71c When we are using an alternative stack for processing SIGSEGV, we cannot execute arbitrary lisp code there -- instead, jump to the outermost prompt. 2009-09-25 18:44:15 +02:00
Juan Jose Garcia Ripoll
c36eb22e73 Stack overflows now jump to the outermost toplevel. 2009-09-25 17:45:34 +02:00
Juan Jose Garcia Ripoll
5f2e473d36 Remove debug statements from last commit 2009-09-25 16:59:36 +02:00
Juan Jose Garcia Ripoll
f8817cda45 SIGSEGV and SIGBUS must be handled synchronously. 2009-09-25 16:58:55 +02:00
Juan Jose Garcia Ripoll
8082cd6010 _ecl_win32_... has to be declared as using WINAPI. 2009-09-25 08:58:34 +02:00
Juan Jose Garcia Ripoll
28f2f9f4ca Export ECL_THREADS configuration flag when processing msvc/ecl/config.h.msvc6 2009-09-25 08:49:04 +02:00
Juan Jose Garcia Ripoll
8edb94230e We have to export _ecl_w32_exception_filter() for users of CL_CATCH_ALL. 2009-09-25 08:42:20 +02:00
Juan Jose Garcia Ripoll
ac8b96db2b We can only wait for the signals that have been blocked, not for SIGUSR1, SIGUSR2, etc. 2009-09-24 19:45:12 +02:00
Juan Jose Garcia Ripoll
bdf8d9d037 Initial version of semaphores both for POSIX and Windows. Includes test that detects failure of unnamed semaphores under Darwin. The whole code is still inactive. 2009-09-23 23:52:47 +02:00
Juan Jose Garcia Ripoll
cc052cb792 Added two new functions mp:lock-mine-p and mp:lock-count-mine which are used to implement mp:with-lock 2009-09-23 22:25:57 +02:00
Juan Jose Garcia Ripoll
9c416fbaa1 wakeup_function() only has to be defined under Windows. 2009-09-23 22:19:40 +02:00
Juan Jose Garcia Ripoll
8c7a40ffe0 The lock count by itself is not enough a guarantee that we acquired the lock
(A. Gavrilov)
2009-09-22 22:37:23 +02:00
Juan Jose Garcia Ripoll
1e12ead719 The routines that wait on condition variables should verify that the
calling thread owns the lock and then mark it as released, at least
for the duration of pthread_count_timedwait().
2009-09-22 22:32:09 +02:00
Juan Jose Garcia Ripoll
a943df3c5f In cl_get_lock(), compute own_process early, so that we do not cause an interrupt
by accessing ecl_env (A. Gavrilov).
2009-09-22 22:18:22 +02:00
Juan Jose Garcia Ripoll
78e9859bd2 Code to implement mp:interrupt-process in Windows:
* unixint.d: capture exceptions caused by page guards.
* unixint.d: to interrupt a Windows thread, set up a page guard onto the
  ECL environment and in addition queue an APC call that accesses the
  environment to trigger that exception.
* stacks.h: CL_CATCH_ALL uses _try/_except to enforce ECL's exception
  handler. Without it, Windows behaves randomly and sometimes uses the
  default handler and sometimes it simply aborts.
* threads.d: we use ordinary handlers to identify threads.
* time.d: SLEEP now uses Windows' SleepEx, leaving the thread in an
  alertable state -- that is, it can be interrupted.
* top.lsp: fixed and improved the code that handles console interrupts,
  by first using ordinary conditions and then using an interactive
  query function to decide which process to interrupt.
2009-09-22 20:13:23 +02:00
Juan Jose Garcia Ripoll
161926ac93 Solved some problem with the queueing of signals and added preliminary code for MP:INTERRUPT-PROCESS using Windows page guards. 2009-09-20 21:53:02 +02:00
Juan Jose Garcia Ripoll
80cf4ea148 Fixed a few typos in the implementation of Windows interrupts. 2009-09-19 00:48:17 +02:00
Juan Jose Garcia Ripoll
2661b663ae New interrupt handling model:
* threads.d: New code to suspend and resume processes.
* threads.d: Threads are created detached with more robust code to wait
  for completion.
* top.lsp: New code for handling console interrupt (Ctrl-C) allows
  selectively interrupting one or more processes or continuing.
* file.d: Console streams are created in non-locking mode when needed.
* load.lsp.in: mp.lsp is always loaded.
* mp.lsp, symbols_list*: without-interrupts and the like belong in EXT
  not in MP package.
2009-09-18 23:02:56 +02:00
Juan Jose Garcia Ripoll
1bba5bab7c Document the rationale behind how we handle interrupts. 2009-09-16 22:30:59 +02:00
Juan Jose Garcia Ripoll
b1ccedea3b Implement ecl_clear_interrupts[_env]() as described in the manual. 2009-09-16 22:30:40 +02:00
Juan Jose Garcia Ripoll
70287ee9c5 Implement missing EXT:WITH-INTERRUPTS 2009-09-16 22:30:16 +02:00
Juan Jose Garcia Ripoll
50c10a5d8d We have to restore the signal before jump_to_sigsegv_handler() 2009-09-16 20:45:15 +02:00
Juan Jose Garcia Ripoll
e647d62b33 unixint.d:
* The code that decides whether to process or queue a signal is factored into
  handle_or_queue. This code is now used by both Linux and Windows handlers.
* Each Linux or Windows handler translates the signal into a lisp object that
  is passed to handle_or_queue. This allows for simpler, platform independent
  queueing, and in particular it enables Windows to discriminate FPE errors.
2009-09-15 23:45:51 +02:00
Juan Jose Garcia Ripoll
90ef6e70ef unixint.d:
* The routine handle_fpe_signal() is embedded in the lisp_signal_handler
* The Windows signal handler falls back to the previously installed handler
  for exceptions that are not to be handled.
2009-09-15 21:59:58 +02:00
Juan Jose Garcia Ripoll
9d49df3a27 unixfsys.d: removed a debug statement that prints the DLL location. 2009-09-15 16:11:51 +02:00
Juan Jose Garcia Ripoll
260853ef0a QUIT now unwinds all forms before exiting. In addition it can kill all existing threads. 2009-09-13 20:46:21 +02:00
Juan Jose Garcia Ripoll
a986a152dd When rebasing the latest changes, the code for with-lock got corrupt. 2009-09-13 20:43:40 +02:00
Juan Jose Garcia Ripoll
a43206af76 Back to using locking streams for I/O in multithreaded environments. 2009-09-13 19:25:44 +02:00
Juan Jose Garcia Ripoll
c4c96f86be Implemented code to allow multiple toplevels sharing the same console and switching from one another. This simplifies the debugger, as the process switching facility and locking is out of that code. 2009-09-13 19:25:44 +02:00
Juan Jose Garcia Ripoll
c61975728a Avoid using C streams for the standard input when using threads. 2009-09-13 19:25:44 +02:00
Juan Jose Garcia Ripoll
ffe97de59c top.lsp: the loop in with-grabbed-console was wrong, for after getting the signal from the condition variable one has to start over again. 2009-09-13 19:25:44 +02:00
Juan Jose Garcia Ripoll
74c8ce560c The console is now protected by a global lock, so that toplevels from different threads do not interfere each other. 2009-09-13 19:25:44 +02:00
Juan Jose Garcia Ripoll
31a2348110 threads.d: fixed typos in the definition of mp_restore_signals. 2009-09-13 19:25:43 +02:00
Juan Jose Garcia Ripoll
2b9b19b669 New functions MP:BLOCK/RESTORE-SIGNALS block signals in a thread-local fashion whenever possible. 2009-09-13 19:25:43 +02:00
Juan Jose Garcia Ripoll
710bfbac2e stacks.h: fixed typo in the definition of CL_CATCH_END 2009-09-13 19:25:43 +02:00
Juan Jose Garcia Ripoll
6d3eda35c4 src/configure.in: detect existence of gmp.h or gmp/gmp.h and signal error when using system libraries without finding the header. 2009-09-09 14:52:14 +02:00
Juan Jose Garcia Ripoll
1cac2fcb4f MP:PROCESS-JOIN was defined also in not threaded builds 2009-09-09 11:58:57 +02:00