Bleadperl v5.15.2-436-gbd31be4 breaks STBEY/Bit-Vector-7.1.tar.gz #11658
Comments
From @andkgit bisect commit bd31be4 Probe for <stdbool.h>, and if found use it in handy.h This means that the core uses the compiler's bool type if one exists. diagnostics cc -c -D_REENTRANT -D_GNU_SOURCE -fno-strict-aliasing -pipe -fstack-protector -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -O2 -g -DVERSION=\"7.1\" -DXS_VERSION=\"7.1\" -fPIC "-I/home/src/perl/repoperls/installed-perls/perl/v5.15.2-436-gbd31be4/2a56/lib/5.15.2/x86_64-linux-thread-multi-ld/CORE" Vector.c perl -V Summary of my perl5 (revision 5 version 15 subversion 2) configuration: Characteristics of this binary (from libperl): -- |
From @andkBuilding Math-Prime-XS -- |
From @nwc10On Sun, Sep 18, 2011 at 04:43:41AM -0700, Andreas J. Koenig via RT wrote:
Yes. As I understand it, from looking at the code: BitVector.c provides the C implementation of bit vectors. BitVector.c always includes exactly 5 headers, <stdlib.h>, <limits.h>, Vector.xs includes "EXTERN.h", "perl.h", "XSUB.h", "patchlevel.h" BitVector.h includes "ToolBox.h" [Curiously, BitVector.c does not include BitVector.h] "ToolBox.h" has this: #ifdef PERL_DARWIN and this: #ifdef __cplusplus PERl_DARWIN will be defined on Darwin, because it's in the CCFLAGS in What goes wrong with the above is when perl.h (indirectly) includes stdbool.h, typedef enum { false = FALSE, true = TRUE } boolean; become syntactically invalid. In turn, the "obvious" fix of changing the "PERL_DARWIN" to It compiles. Regression tests fail. For no obvious reason. It turns out that what is going wrong is, well, quite wonderful: $ gdb --args `perl -lne 'print $1 if /^PERL = (.*)/' Makefile` -Mblib -MBit::Vector -lwe '($v0, $v1, $v2) = map {Bit::Vector->new(31)} 0..2; $v1->from_Hex("FFFFFFFE"); $v2->from_Hex("FFFFFFFE"); $v0->add($v1, $v2, 0); print $v0->to_Hex();' Breakpoint 1 (BitVector_compute) pending. Breakpoint 1, BitVector_compute (X=0x9b529c, Y=0x9c1b8c, Z=0x9c377c, In the two different files, the type boolean ends up having a different size. Took a bit of time to figure that one out. After a lot of goes at trying to find an elegant fix, this seems to be it: Inline Patch--- Vector.xs~ 2009-09-29 22:54:00.000000000 +0100
+++ Vector.xs 2011-09-22 16:20:21.000000000 +0100
@@ -10,6 +10,7 @@
/* */
/*****************************************************************************/
+#include "BitVector.h"
#include "EXTERN.h"
#include "perl.h"
@@ -30,10 +31,6 @@
#endif
#endif
-
-#include "BitVector.h"
-
-
typedef SV *BitVector_Object;
typedef SV *BitVector_Handle;
typedef N_word *BitVector_Address;
(No compiler warnings or errors - everything else I tried seemed to have This does feel fragile to me. Nicholas Clark |
|
The RT System itself - Status changed from 'new' to 'open' |
From @nwc10On Tue, Sep 20, 2011 at 08:10:00AM +0200, Andreas J. Koenig wrote:
It reached some archives, eg http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2011-09/msg00810.html
Reply to STBEY sent as a reply to the original message. Fix for Math-Prime-XS seems much simpler: Inline Patch--- lib/Math/Prime/XS.xs~ 2011-09-22 11:36:46.000000000 +0100
+++ lib/Math/Prime/XS.xs 2011-09-22 11:40:09.000000000 +0100
@@ -21,7 +21,9 @@
#define NUM_LEN(nums) (sizeof (nums) / sizeof (num_entry))
+#ifndef I_STDBOOL
enum { false, true };
+#endif
typedef struct {
unsigned long **ptr;
Nicholas Clark |
From zefram@fysh.orgNicholas Clark wrote:
Better fix: in the header change typedef enum { false = FALSE, true = TRUE } boolean; to typedef int boolean; This means that the type of "boolean" is consistent per system, and the -zefram |
From zefram@fysh.orgNicholas Clark wrote:
That formulation could also be used in the critical part of the fix that -zefram |
From @andk
> Were these the only two distributions to have problems with perl.h I can't tell what my current coverage is. Usually cpantesters and -- |
From schubiger@cpan.orgNicholas Clark <nick@ccl4.org> wrote:
Thanks to both for reporting and the diagnosis. Fixed in v0.23_01. |
From @andk
> It reached some archives, eg > http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2011-09/msg00810.html Ahh, interesting, I just retried searching for it by number: http://www.xray.mpe.mpg.de/cgi-bin/w3glimpse/perl5-porters?query=99408&errors=0&case=on&maxfiles=100&maxlines=30 Gives 0 matches. Many other queries that should find it fail as well. -- |
From @rjbsI cannot reproduce failures for this on my Darwin box. Is this issue still blocking? Steps to reproduce, please, if so. |
From @jkeenanOn Tue Feb 28 17:26:39 2012, rjbs wrote:
Along the same lines, if: (a) I have installed Bit-Vector version 7.1 on my machine; (b) I have retained the Bit-Vector test suite under (c) I have a fairly up-to-date, tested version of blead perl under How would I run the test suite with blead? In particular, what would I ##### (I suspect that, more generally, this comes down to: How do I test Thank you very much. |
From @rjbs* James E Keenan via RT <perlbug-followup@perl.org> [2012-02-28T21:10:03]
Rather than answer just what you asked, here's what I do. I use perlbrew to keep the latest 5.15.x installed, as well as blead, when So, say I want to test Bit::Vector. I run: perlbrew lib create 5.15.8@bit-vector ...to create a "bit-vector" local::lib attached to my "5.15.8" perlbrew. Then: cpanm --look Bit::Vector ...which drops me in a shell with the latest Bit::Vector unpacked. Then: perl Makefile.PL && make test And that's about it. -- |
From @andk
> I cannot reproduce failures for this on my Darwin box. I just take four recent bleadperls: - With all four I run 'install Bit::Vector' on my cpan shell and get: make[1]: Entering directory `/tmp/tmp.hg0fEzbF6R/Bit-Vector-7.1-HPmcsK' CPAN::Reporter: this appears to be a duplicate report for the make phase: Test report will not be sent. STBEY/Bit-Vector-7.1.tar.gz Patch is in ticket https://rt.cpan.org/Ticket/Display.html?id=71052 Also in this ticket Steffen Beyer has authorized me to make a new -- |
From @rjbs* "Andreas J. Koenig" <andreas.koenig.7os6VVqR@franz.ak.mind.de> [2012-02-29T02:02:49]
Huh! Thanks for testing that.
...and thanks even more for that! Once it's out and the fix confirmed against -- |
From @cpansproutOn Wed Feb 29 06:30:01 2012, perl.p5p@rjbs.manxome.org wrote:
Steffen Beyer mentioned in that ticket that he doesn’t have time to make -- Father Chrysostomos |
From @rjbs* Father Chrysostomos via RT <perlbug-followup@perl.org> [2012-02-29T12:16:49]
Andreas K., above, said he would be making it. -- |
From @rjbsI spoke with David W. via email and he says he will cut a new release very soon. |
From @rjbsOn Wed Feb 29 18:43:08 2012, rjbs wrote:
Whoops, that comment belongs on [rt.perl.org #109806] |
From @andk
> * Father Chrysostomos via RT <perlbug-followup@perl.org> [2012-02-29T12:16:49]
> Andreas K., above, said he would be making it. The uploaded file Bit-Vector-7.2.tar.gz has entered CPAN as file: $CPAN/authors/id/A/AN/ANDK/Bit-Vector-7.2.tar.gz Ticket can be closed -- |
From @rjbsThanks, Andreas! Closed |
From [Unknown Contact. See original ticket]Thanks, Andreas! Closed |
|
@rjbs - Status changed from 'open' to 'resolved' |
Migrated from rt.perl.org#99408 (status was 'resolved')
Searchable as RT99408$
The text was updated successfully, but these errors were encountered: