freenode/#sbcl - IRC Chatlog
Search
1:40:51
stassats
thinking about specialized structure arrays, the hardest part would be deciding what to do when escaping a value
1:45:13
stassats
that greatly simplifies the implementation, just need to solve for interior pointers
1:52:38
nyef``
Include some flag and backpointer on the structure header to point to the array header.
12:13:30
scymtym
stassats: re TYPECASE: i think something like the decision tree generator i made for the fastgf experiment could generate good code. it loads widetags and layouts only when needed and only once and does the minimum number of tests by sharing information between clauses. it would probably have to punt for non-structure classes since we wouldn't want to invoke the compiler from TYPECASE
12:14:59
scymtym
sure, i was thinking about identifying shapes like (if (typep …) (if typep …) (if typep …))
12:17:16
attila_lendvai
re ASDF and "updating to the newer API so that it breaks on older setups"... ASDF has its upgrade functionality, to help to "escape forward". I don't see any reasons to run with old ASDF when all it takes is to check out a fresh version somewhere and then (asdf:load-system :asdf) to initiate an upgrade, even if your lisp ships and older version. I've been running with that setup since runtime upgrade became a thing.
12:18:59
scymtym
the problem seems to be that on the x86 windows slave (apparently not on the x86_64 one), the sb-concurrency test hangs or spins in way that makes the process stick around. jenkins cannot terminate it and some files stay in use. this prevents the workspace from being deleted for the next build (and ties up a cpu on the vm)
12:19:35
scymtym
otherwise the windows builds are remarkably reliable. without running tests, to be fair
12:20:14
Shinmera
scymtym: Does it happen even if you just build on the x86_64 machine with the x86 msys2?
12:57:46
stassats
scymtym: i can identify (if (typep …) (if typep …) (if typep …)) trivially, but then i can't do anything
12:58:27
stassats
so, a let clambda needs to be introduced, wrapping some arbitrary blocks, which blocks? etc.
13:01:48
scymtym
stassats: the problems are on the x86 win32 vm, the x86_64 win32 vm seems fine so far
13:02:17
stassats
there isn't even a good spec to adhere to when writing .asd files, so you're subject to some new retrospective decisions
13:04:17
scymtym
well, the syntax and semantics are discussed in the manual. the "best practices" thing is complementary
13:07:18
scymtym
i agree that modern ASDF makes using dependencies, in particular not enthusiastically maintained ones, a lot more painful