freenode/#sbcl - IRC Chatlog
Search
8:58:53
scymtym
dougk: lots of crashes "GC invariant lost, file "marknsweepgc.c", line 2259" on x86_64 when dumping an image. seems to have started with commit "x86-64: Save absolute assembler routine fixups"
11:38:02
scymtym
there is also this macos build failure since bf2e122e: http://paste.lisp.org/display/355643
14:39:22
dougk_
ok, i'll take a look at the GC lossage. I'm not having any build failure on macOS, either 32 or 64-bit
14:42:12
dougk_
scymtym: tell me your build features for the failure; stassats: you're right, i messed up that thing. I might just add simple-fun headers to assembly code. But not necessary since I do know the range of assembly code components.
18:10:23
|3b|`
looks like 27f36d4decfe336b95c8ae63c12ac4c5959671f3 "Impart consistency to accessing thread-local Lisp vars from C"
18:12:39
|3b|`
possibly due to SetSymbolValue that didn't get renamed to write_TLS inside #ifdef LISP_FEATURE_WIN32
18:13:09
dougk
|3b|`: correct, can you try changing to "write_TLS(THRUPTION_PENDING, T, p)" and let me know ?
18:16:29
dougk
ah, see, so my commit message was right - sometimes people used "SetSymbolValue" when they meant "SetTlSymbolValue" and vv, but they're really the same thing; and I got confused trying to unconfuse it
18:28:06
dougk
I attributed the fix to you. It didn't fail for me as I only test on linux, and by "test" I mean compile, never actually run w/safepoint
18:54:29
dougk_
cold-init is able to succeed on x86-64 with both immobile and dynamic spaces mapped at "wrong" addresses now.
19:28:42
scymtym
dougk_: the macos build fails for all configurations: e.g. https://ci.cor-lab.org/job/sbcl-master/2444/ . maybe this is just our slave running old software again
19:33:16
dougk_
there's no (extern-alien ) in make-target-2-load. Any chance you can see what happens if you #+nil out the call to shake-packages?