0000165: Re: a few practical details

Notes
(0000422) administrator (administrator) 2000-07-25 18:40	> - ocamldebug uses $PATH to find the prog to debug. This is quite > annoying if you don't have ``.'' in your path (as it should be :) Yes, that was silly. It's fixed in the working sources. > As for now, i just patched camldebug.el to workaround this (as > anyway, ocamldebug can't be used standalone on command line with no > readline! By the way what about adding readline support to ocaml and > ocamldebug?) Initially, we hoped that an external line editor such as Ledit could be used. However, external line editors aren't widespread nowadays. > - could it be added to the documentation that ocamldebug and > ocamlbrowser should not be stripped, unless they don't work the > way they should (the worse is maybe that they work just like > ocamlrun, making you go crazy trying to find out what the hell is > going on, moreover together with the PATH pb!) Right. All programs compiled with "ocamlc -custom" must not be stripped. This is mentioned in the docs, but the installation instructions should also say it for ocamldebug and ocamlbrowser. > - are there any good solutions to debug tk-enabled programs? As for > now, i managed it by adding the different librairies i'm using to > compilation of ocamldebug, need the source at hand though. The > second problem is of course that back debugging doesn't work (you > get Xlib: unexpected async reply (sequence 0x6a8)!), so sad when > you're already addicted to it! Right. I'm afraid reverse execution will never work for programs that do I/O, especially socket I/O, and especially socket I/O done in an external C library... > - lastly, a bug (is it?) on axp: > > >> pixel@alpha:~>ocaml > >> Objective Caml version 3.00 > >> > >> # 1.0 /. 0. ;; > >> zsh: floating point exception ocaml > > whereas on ix86 & sparc it gives: > > >> pixel@leia:~>ocaml > >> Objective Caml version 3.00 > >> > >> # 1.0 /. 0.0 ;; > >> - : float = inf Well, you'd get the same floating-point exception on an Alpha with a C program. The Alpha processor doesn't implement all of IEEE arithmetic in hardware, in particular the treatment of infinities. Since the OCaml virtual machine is in C, you get the same behavior in OCaml. This said, there are special flags to the C compiler that allow full IEEE support via software assistance (-mieee? -fieee? can't remember). I think we should activate those flags when compiling the bytecode interpreter, at least. (The native-code compiler is another story.) Thanks for your feedback, - Xavier Leroy

(0000423) administrator (administrator) 2000-07-26 00:16	Xavier Leroy <Xavier.Leroy@inria.fr> writes: [...] > > - could it be added to the documentation that ocamldebug and > > ocamlbrowser should not be stripped, unless they don't work the > > way they should (the worse is maybe that they work just like > > ocamlrun, making you go crazy trying to find out what the hell is > > going on, moreover together with the PATH pb!) > > Right. All programs compiled with "ocamlc -custom" must not be > stripped. This is mentioned in the docs, but the installation > instructions should also say it for ocamldebug and ocamlbrowser. ?? i've done a "grep -r -i strip ." in the ocaml source directory and didn't find any mention about this [...] > > >> pixel@alpha:~>ocaml > > >> Objective Caml version 3.00 > > >> > > >> # 1.0 /. 0. ;; > > >> zsh: floating point exception ocaml [...] > Well, you'd get the same floating-point exception on an Alpha with a C > program. The Alpha processor doesn't implement all of IEEE arithmetic > in hardware, in particular the treatment of infinities. Since the > OCaml virtual machine is in C, you get the same behavior in OCaml. > > This said, there are special flags to the C compiler that allow full > IEEE support via software assistance (-mieee? -fieee? can't remember). ok, i've rebuilded ocaml with -mieee, and now it works nicely. the pb i have now is to have the native-code compiler work the same... pixel@alpha:/tmp>echo "1. /. 0." > t.ml ; ocamlc t.ml ; ./a.out pixel@alpha:/tmp>echo "1. /. 0." > t.ml ; ocamlopt t.ml ; ./a.out zsh: floating point exception ./a.out i've tried passing some -mieee to ocamlopt, but as ocamlopt generates assembler, i had no luck of course... as for now, i'll keep the bytecode version which works :) thanks, cu Pixel.

(0000424) administrator (administrator) 2000-07-27 11:32	> > Right. All programs compiled with "ocamlc -custom" must not be > > stripped. This is mentioned in the docs, but the installation > > instructions should also say it for ocamldebug and ocamlbrowser. > > ?? i've done a "grep -r -i strip ." in the ocaml source directory and didn't > find any mention about this The manual is distributed separately from the sources. > ok, i've rebuilded ocaml with -mieee, and now it works nicely. > the pb i have now is to have the native-code compiler work the same... > i've tried passing some -mieee to ocamlopt, but as ocamlopt > generates assembler, i had no luck of course... Right. It would be necessary to modify the code generator, so that it generates enough hooks for the software fix-up code to work, and it's not obvious at all how to do it. - Xavier Leroy

(0000425) administrator (administrator) 2000-07-27 13:20	> ?? i've done a "grep -r -i strip ." in the ocaml source directory and didn't > find any mention about this Sorry, I mis-remembered: the "strip" trap wasn't documented in the manual either. Now, it is. Thanks for your feedback. - Xavier Leroy

(0000426) administrator (administrator) 2000-07-27 13:34	Debugger problem fixed some time ago. INSTALL file, docs, and Alpha compilation options fixed 2000-07-27 by XL.

Issue History
Date Modified	Username	Field	Change
2005-11-18 10:13	administrator	New Issue

Relationships