Bug 204978 - lang/php56: zend_strtod.c needs __VFP_FP__ defined to work correctly on arm AT91SAM9
Summary: lang/php56: zend_strtod.c needs __VFP_FP__ defined to work correctly on arm ...
Status: Closed Overcome By Events
Alias: None
Product: Ports & Packages
Classification: Unclassified
Component: Individual Port(s) (show other bugs)
Version: Latest
Hardware: arm Any
: --- Affects Some People
Assignee: Alex Dupre
URL:
Keywords: needs-patch, needs-qa
Depends on:
Blocks:
 
Reported: 2015-12-02 22:17 UTC by David Horwitt
Modified: 2019-01-01 13:43 UTC (History)
6 users (show)

See Also:
koobs: maintainer-feedback? (ale)
koobs: merge-quarterly?

Attachments
Full output of "cc -E -dM - </dev/null" (4.66 KB, text/plain)
2015-12-04 15:18 UTC, David Horwitt 		no flags Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description David Horwitt 2015-12-02 22:17:02 UTC 
<?php
session_start();
?>

This fails with "PHP Fatal error:  Balloc() allocation exceeds list boundary in - on line 2"

This seems to trace from zend_dtoa() in Zend/zend_strtod.c.
#define __VFP_FP__ before line 156 seems to resolve the error.
Comment 2 David Horwitt 2015-12-03 05:32:45 UTC 
After applying patch the failing behavior is the same; explicitly defining
__VFP_FP__ restores correct behavior.

It looks like the patch asumes that __VFP_FP__ is supplied by clang (builtin) which isn't the case on my system.
Comment 3 Michael Moll freebsd_committer freebsd_triage 2015-12-03 14:21:58 UTC 
is this on armv6 or armv6hf?
Comment 4 David Horwitt 2015-12-03 18:42:23 UTC 
AT91SAM9 = ARM926EJS = ARMv5TEJ.

I think; ARM nomenclature hurts my brain.

My buildworld and buildkernel used TARGET=arm and TARGET_ARCH=arm (cross-built on i386).
Comment 5 Mikael Urankar freebsd_committer freebsd_triage 2015-12-04 14:55:02 UTC 
What does "cc -E -dM - < /dev/null | grep VFP_FP" returns?
Comment 6 David Horwitt 2015-12-04 15:18:20 UTC 
Created attachment 163846 [details]
Full output of "cc -E -dM - </dev/null"
Comment 7 David Horwitt 2015-12-04 15:20:14 UTC 
More direct reply to comment 5:

Script started on Fri Dec  4 14:12:46 2015
# uname -v
FreeBSD 10.2-STABLE #1: Tue Nov  3 20:18:10 UTC 2015     root@dstar2:/data/src/sys/arm/compile/SOM9G20M 
# cc -E -dM - </dev/null | grep VFP_FP
# exit
exit

Script done on Fri Dec  4 14:13:10 2015
Comment 8 Michael Moll freebsd_committer freebsd_triage 2015-12-04 16:22:08 UTC 
David, could you try if it works when you just define IEEE_LITTLE_ENDIAN?
Comment 9 David Horwitt 2015-12-04 18:00:05 UTC 
Does _not_ work when IEEE_LITTLE_ENDIAN is defined.

NB:
 1) Using original zend_strtod.c (patch in comment 1 not applied)
 2) Tried defining IEEE_LITTLE_ENDIAN in various places
    a) at line 155
    b) at line 285
Comment 10 Michael Moll freebsd_committer freebsd_triage 2015-12-06 12:52:49 UTC 
According to #bsdmips none of the ARMv5 processors we do support with the "arm" architecture does have VFP, so this might be just an accidental fix by some cross effects. I does work for me on armv6hf and I'm quite sure it would also work on armv6.

I tried to tinker around on my old Sheevaplug to get more insights, but it seems clang is broken on -CURRENT (see https://lists.freebsd.org/pipermail/freebsd-arm/2015-October/012491.html ff.).

-> assigning ticket back to ports pool
Comment 11 Kubilay Kocak freebsd_committer freebsd_triage 2015-12-24 18:13:02 UTC 
Correctly assign to MAINTAINER
Comment 12 Walter Schwarzenfeld 2018-01-12 13:07:56 UTC 
Maintainer feedback?
Comment 13 Rene Ladan freebsd_committer freebsd_triage 2019-01-01 13:43:02 UTC 
Expired port removed, no recent updates.