Poly/ML

polyml@lists.polyml.org

2 participants
698 discussions

[isabelle-dev] PolyML crashes
by michael.norrish＠nicta.com.au 03 Mar '14

03 Mar '14

I seem to be able to pretty reliably produce the following: Exception- InternalError: indirect - invalid constant address raised while compiling Exception- InternalError: indirect - invalid constant address raised while compiling Unfortunately, it takes hours to get there... Switching to --gcthreads=1 may have fixed another error we were having, relating to unwarranted pointer-equality tests. Michael On 21/02/14 22:30, David Matthews wrote: > Hi, > I'm aware of a number of assertion failures that seem to occur intermittently. > This is one of the ones on my list. I suspect they are all symptoms of the same > bug but I have never been able to narrow it down or reproduce it. It does seem > to occur when the memory management is under heavy load. > > Thanks for reporting it and if anyone manages to reproduce it more consistently > I would be very interested. > > Regards, > David > > On 20/02/2014 14:21, Ond?ej Kun?ar wrote: >> Hi! >> In the past couple of months I've gotten a crash of PolyML always with >> the same error message. I cannot reproduce the problem reliably but >> because it has already happened, let say, six times in the past three >> months, I am reporting the problem here: >> >> Unofficial version of Isabelle/HOL (unidentified repository version) >> poly: gc_mark_phase.cpp:432: virtual void >> MTGCProcessMarkPointers::ScanAddressesInObject(PolyObject*, >> POLYUNSIGNED): Assertion `baseAddr > (PolyWord*)obj && baseAddr < >> ((PolyWord*)obj)+length' failed. >> >> >> This refers to any changeset in the past three months. >> >> Best, >> Ondrej >> _______________________________________________ >> isabelle-dev mailing list >> isabelle-dev at in.tum.de >> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev > _______________________________________________ > isabelle-dev mailing list > isabelle-dev at in.tum.de > https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

3 3

(Pseudo) static linking with C
by list.2006＠nul.jp 19 Feb '14

19 Feb '14

Hi, Although Poly/ML has nice interface for calling C functions, it only supports dynamic linking. <http://www.polyml.org/docs/CInterface.html> This is not pleasant if you want to distribute a program as a simple stand-alone executable. (setting the library path might be crumsy for the user) A solution would be 'linking statically but loading dynamically.' Actually, dlopen seems to support this if the filename argument is NULL. The following experiment (on Ubuntu 13.04 x86_64) seems to have succeeded, but needs a little modification to Poly/ML. How do you think? Any chance for Poly/ML to incorporate something similar? Best regards, -- KOBAYASHI, Tomoyas ===== patch ===== --- foreign.cpp.orig 2014-02-15 00:29:32.334802335 +0900 +++ foreign.cpp 2014-02-15 00:35:58.362818505 +0900 @@ -742,7 +742,7 @@ return res; #else /* UNIX version */ - void *lib = dlopen(name,DLOPENFLAGS); + void *lib = dlopen(*name == 0 ? NULL : name,DLOPENFLAGS); if (!lib) { char buf[256]; ===== experiment session ===== $ cat difference.c int difference (int x, int y) { return x > y ? x - y : y - x; } $ gcc -c difference.c -fPIC -o difference.o $ cat diff.sml local open CInterface in fun diff (n, m) = let val lib = load_lib "" val sym = load_sym lib "difference" val diff' = call2 sym (INT, INT) INT in diff' (n, m) end fun main () = print (Int.toString (diff (13, 50)) ^ "\n"); end $ poly Poly/ML 5.5.2 Testing > use "diff.sml"; val diff = fn: int * int -> int val main = fn: unit -> unit val it = (): unit > PolyML.export("diff.o", main); val it = (): unit > $ gcc -L ~/poly/lib -rdynamic diff.o difference.o -lpolymain -lpolyml -lstdc++ -lm -ldl -lpthread -lgmp $ ./a.out 37 $

2 2

Subsecond time in stat structures
by loganjerry＠gmail.com 18 Feb '14

18 Feb '14

I would like to propose a patch to PolyML (attached) that uses subsecond time fields in stat structures, when they are available. I think the patch is fairly straightforward. When I changed configure.ac by adding the AC_CHECK_MEMBERS macro, autoconf started whining at me about AC_CANONICAL_HOST being obsolete. Then it turned out that its replacement, AC_CANONICAL_TARGET, must appear before AM_INIT_AUTOMAKE, hence the slight rearrangement of macro calls there. Ignore that part if you like AC_CANONICAL_HOST. I am not sure of the best place to put the STAT_SECS and STAT_USECS macros. They seemed to me to be somewhat related to other definitions in io_internal.h, but perhaps there is a better place to put them. Anyway, this is more a proof of concept than a finished product. Feel free to mangle it to your heart's content. I won't be offended. :-) Regards, -- Jerry James http://www.jamezone.org/

2 2

Exception- Subscript unexpectedly raised while compiling
by noschinl＠in.tum.de 17 Feb '14

17 Feb '14

Hi everyone, today I stumbled upon some strange behaviour of PolyML 5.5.1 (the x86-linux version shipped with Isabelle). The behaviour also occurred with revision 1874 from the SVN (Sourceforge has problems at the moment, so I didn't test the most current version). ------------------------------------------------------ # ./poly Poly/ML 5.5.1 Release > fun f xs = (xs : ''a) val _ = (fn x => f x); Exception- Subscript unexpectedly raised while compiling Exception- Subscript unexpectedly raised while compiling ------------------------------------------------------ But: ------------------------------------------------------ # ./poly Poly/ML 5.5.1 Release > fun f xs = (xs : ''a); val _ = (fn x => f x); val f = fn: ''a -> ''a ------------------------------------------------------ But: ------------------------------------------------------ # ./poly Poly/ML 5.5.1 Release > let fun f xs = (xs : ''a); val _ = (fn x => f x) in () end; Exception- Subscript unexpectedly raised while compiling Exception- Subscript unexpectedly raised while compiling ------------------------------------------------------ So the error vanishes, if I add a semicolon before the "val", except when I am in a "let" block. Moreover, the error seems to depend on the equality type and the dummy pattern. Best regards, Lars

2 2

printInAlphabeticalOrder
by rda＠lemma-one.com 13 Feb '14

13 Feb '14

I am confused about what PolyML.Compiler.printInAlphabeticalOrder is intended to control. If I enter the following: signature S = sig val x : int val z : int val y : int end; structure S = struct val x = 1 val z = 3 val y = 2 end; the signature and the structure are both printed with the variables in reverse alphabetical order. (I was rather hoping for an option to have them in the source code order.) Regards, Rob.

2 1

Pretty-printing infix constructors
by Rda＠lemma-one.com 06 Feb '14

06 Feb '14

When I executed the following infix xxx; datatype XXX = op xxx of int * string; 1 xxx "banana"; I was rather hopping the expression at the end would print out with infix notation but it doesn't: val it = xxx (1, "banana"): XXX Is there an option in Poly/ML to have infix constructors print as infixes, or do I need to write my own pretty-printer to get that. Regards, Rob.

2 1

PolyML.pointerEq
by Brian.Campbell＠ed.ac.uk 05 Feb '14

05 Feb '14

What semantics, if any, does PolyML.pointerEq have? In particular, is it supposed to be guaranteed that if pointerEq (x,y) is true, then it will remain true after GC? My reason for asking is that I've been having trouble with assertion failures in HOL when reducing large terms. The assertions check that the left hand side of the conclusions of two theorem objects are equal according to pointerEq, because this allows a fast comparison that "should" always be true if the theorem is manipulated carefully. I'm fairly certain the assertion failures are related to GC: they are non-deterministic on code that ought to be deterministic, and vary in frequency across different machines. (I haven't seen it at all on a 4GB 32-bit machine, rarely on 4GB 64-bit, often on the 8GB I normally use and infrequently on a compute server with much more.) There's no sign of any data corruption, however. Brian

3 4

checking --enable-shared from within the REPL
by David.Matthews＠prolingua.co.uk 14 Jan '14

14 Jan '14

On 13/01/2014 22:31, Michael Norrish wrote: > I'll look at the relevant lib directories; thanks. > > The reason is that the HOL build process wants to create executables (for > mlyacc, mllex and others), and doing so without --enable-shared will fail > mysteriously. If I can cut this off at the pass, I can provide better error > messages. There was a problem that I discovered with making --disable-shared the default that I hadn't anticipated. It seems that when linking with the static library it is necessary to include some extra libraries (e.g. gmp) that are used within libpolyml. These don't seem to be needed when linking with the dynamic library so presumably they have already been linked into the dynamic library. > I confess I haven't properly looked at polyc; would this give us a uniform approach? That is the intention. "polyc" is written as part of the process of building "poly" and should include all the libraries that are needed. Regards, David

1 0

checking --enable-shared from within the REPL
by michael.norrish＠nicta.com.au 13 Jan '14

13 Jan '14

Is there any way to detect if the system has been built with --enable-shared? Something in the PolyML structure (say)? I'd like to be able to have my build script abort if the PolyML hasn't been built correctly. Or is there anything obvious my script could check in the file system to make this determination? Thanks, Michael

3 2

excessive compile times?
by Michael.Norrish＠nicta.com.au 10 Dec '13

10 Dec '13

A minimally modified version of the file at https://raw.github.com/mn200/HOL/master/src/bool/boolScript.sml takes 83 seconds to be ?used? (using PolyML.use) into an interactive PolyML session. There are a great many files that need to be used before boolScript gets to work, but none of those files need any longer than 2seconds to be used, and the vast majority take less than a second. The "minimal modifications? turn the ``?`` syntax into valid SML, such that ``p /\ q`` turns into (Term [QUOTE ?p /\\ q?]) and turns `p /\ q` into [QUOTE ?p /\\ q?] We have a timer that starts when the script file actually begins to execute, and this reports that the execution time is less than a second, so it really does seem as if the compilation time is the issue. We have many other Script files in HOL4 that are at least this long, but they don?t have slow compilation times in the same way. Also, the much smaller (30 times by LoC) https://raw.github.com/mn200/HOL/master/src/1/ConseqConvScript.sml has the same prefix as boolScript, but compiles in less than a second. Is there anything obvious (and remediable) that boolScript is doing that might be causing this problem? FWIW, Moscow ML compiles this file as quickly as all the others in the system. Thanks, Michael ________________________________ The information in this e-mail may be confidential and subject to legal professional privilege and/or copyright. National ICT Australia Limited accepts no liability for any damage caused by this email or its attachments.

2 2

← Newer
1
...
26
27
28
29
30
31
32
...
70
Older →

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

Poly/ML