Re RTLD_LAZY, much appreciated.
Re ⎕NC, that's the advantage of some nut writing obscure code: it
identifies obscure bugs.
Henrik
On 10/25/24 08:13, Dr. Jürgen Sauermann wrote:
Hi Henrik,
thanks for the reminder. Fixed in *SVN 1783*.
I also spotted an error which caused ⎕NC to return a wrong value for
⍺, ⍵, etc.
⍺ and ⍵ came after ⎕NC was first implemented, and I failed to handle
them properly.
Note also that ⎕NC is slightly different from IBM's by having
different values for
user defined and system defined names. Therefore ⍺ has a ⎕NC of 5 if
defined.
For example:
* MINUS←{ (⍺-⍵)⊣⍎"0"⊢[5≠⎕NC'⍺']"⍺←0" }
5 MINUS 1
4
MINUS 1
¯1*
Best Regards,
Jürgen
On 10/24/24 18:15, Henrik Moller wrote:
Hey, Jürgen,
Thanks.
This is part of continuing work on the mtx native function thing I
put up six months ago, and just for the sake of consistency, I was
trying make all the mtx functions lambdas, but it's no problem to use
a ∇ function when I need ambivalence. (If I'd looked at how you
implement lambdas and conditionals, I would have known the lambda
approach wouldn't work...)
Speaking of mtx, per the note I sent you a couple days ago, have you
had a chance to look into making the dlopen() in
NativeFunction::try_one_file() use RTLD_LAZY instead of RTLD_NOW? I
don't think anyone but me is using mtx, so it's cool if LAZY isn't in
your plans--it just means I patch the line when you put out a new SVN.
Thanks,
Henrik
On 10/24/24 11:30, Dr. Jürgen Sauermann wrote:
Hi Henrik,
I believe I fixed the Segfault. You should now get
a DEFN ERROR instead with )MORE information.
SVN *1782*.
The reason is this:
1, Conditionals, i.e. COND →→ IFSTAT ←→ ELSESTAT ←←
create 2 or 3 statements, one for COND, one for IFSTAT and
optionally one for ELSESTAT.
2. GNU APL lambdas are limited to one APL expression (= one APL
statement)
and the expression must yield a value (and can therefore not be a
branch).
3. The normal work-around for this limitation is to use a proper
∇-defined function.
Best Regards,
Jürgen
