freenode/#sbcl - IRC Chatlog
Search
12:44:43
Xof
if it's not a regression relative to 1.4.6, and no-one reported it this month, I think it can wait until after the release
12:46:54
scymtym
ok. an epic debugging journey lead me to its discovery, so yes, it is pretty subtle and unlikely to affect anybody else
12:52:40
Xof
it's a nice one, and it maybe suggests that those predicates shouldn't be complete predicates, but only defined on CTYPE structures
12:53:15
Xof
i.e. the structure predicate should be (%intersection-type-p ...), and (defun intersection-type-p (x) (declare (type ctype x)) (%intersection-type-p x))
13:32:47
pfdietz
I specfically added a (the (satisfies eval) ...) mutator because it was showing up in reduced test cases stassats was producing from the bug reports.
13:34:38
pfdietz
On my to-do list is expanding the random-type-prop tests to generate a wider variety of code, including such forms.
13:36:49
pfdietz
Those tests (1000+ of them) encode in a sort-of declarative form information about what valid inputs are to various lisp functions. That should be useful for more that what's it's being used for right now.
13:39:24
pfdietz
I've been mining lambda expressions (and forms to make lambda expressions) from various lisp sources, as grist for mutation testing.
13:40:08
pfdietz
So, I hacked the SBCL reader to have an option so that if it finds an unknown package when parsing a symbol, it just puts t in the current package.
13:42:00
pfdietz
I'm getting about a 250K distinct lambdas from quicklisp, ansi-tests, various github repos, etc.
14:49:10
dougk_
Xof: would be opposed to my pushing an improved fix for https://bugs.launchpad.net/sbcl/+bug/1366263 without which it's not possible to even reliably run cheneygc? I think it's gotten worse somehow
14:55:45
Xof
the analysis in that bug is plausible, but it would be nice to know why it's suddenly worse
14:55:48
dougk_
dont know the timeframe, but i have near identical places at which failure occurs on both emulators and physical hardware
14:56:42
Xof
If it's not new, I'd rather see the patch at the start of 1.4.7.x, unless it only touches cheneygc-only source
14:57:27
Xof
(and even then, I'd like to think twice about committing during the freeze; the thought of possibly invalidating all the testing done so far is a bit miserable)
14:58:13
Xof
I was singing the praises of this attitude to Ludovic Cortes at ELS; I'd quite like to act as if it is true :)
15:01:00
dougk_
so out of curiosity, why can an octet buffer straddle the unprotected/protected boundary? i thought that the lowest protected page strictly bounds the range below which is in-use and above which is not-in-use ?
17:18:20
dougk_
stassats: i bisected the Alpha failure to "Fix stack exhaustion in with-alien on non-x86oids."
20:38:18
phoe
the code walker in SBCL is very aggressive when macroexpanding. when I try to slime-macrostep the external DEFUN in the form (defun foo () (when nil (defun bar 2))) then I get thrown into the debugger.
20:39:08
phoe
I'd find it more convenient to be thrown into the debugger when the actually wrong form is attempted to be macrostepped - if I try to step into a complicated macro, I instantly get an error, instead of being able to expand the outer macro and try expanding inner ones to check for the exact location of the error.