bug#22049: libreoffice compile error

[Leo Famulari <leo@famulari.name>]

[-- Attachment #1: Type: text/plain, Size: 1513 bytes --]

On Mon, Feb 29, 2016 at 03:35:40AM -0500, Leo Famulari wrote:
> On Sun, Feb 28, 2016 at 07:06:22PM +0100, Ludovic Courtès wrote:
> > Leo Famulari <leo@famulari.name> skribis:
> > > It looks like the problem is with a libreoffice dependency, ilmbase.
> > > Looking at the build history on hydra.gnu.org, it hasn't built
> > > successfully since at least 2015-02-25:
> > > http://hydra.gnu.org/job/gnu/master/ilmbase-2.2.0.i686-linux
> > 
> > The problem appears to be fixed in IlmBase 2.0.1:
> > 
> >   https://github.com/openexr/openexr/issues/3
> > 
> > Would you like to try?
> 
> We currently package 2.2.0. I did read that bug report, and then kept
> reading, until I came to the Debian bug report [0], where Steven
> Chamberlain linked to our bug report with OpenEXR and produced two
> patches that Debian has merged into their "unstable" branch.
> 
> I've attached my (working) attempt at conditionally applying these
> patches on i686-* architectures. Please feel free to criticize or modify 
> this patch :)

I forgot the attachment. Here it is!

> 
> I copied the patches from [1].
> 
> Thanks for the reminder!
> 
> [0]
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815712
> 
> [1]
> https://sources.debian.net/src/ilmbase/2.2.0-11/debian/patches/
> 
> [2]
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815712#194
> 
> Relevant commit in Debian ilmbase packaging repo:
> https://anonscm.debian.org/cgit/pkg-phototools/ilmbase.git/commit/?id=ab28bb45cdad8adc32e345b777ab8e692b1d9a9c
> 
> 
> 

[-- Attachment #2: 0001-gnu-ilmbase-Add-patches-to-fix-build-on-i686.patch --]
[-- Type: text/x-diff, Size: 9052 bytes --]

From bf1f2a1c3621ba93ec99711ec78a79663acb6e82 Mon Sep 17 00:00:00 2001
Message-Id: <bf1f2a1c3621ba93ec99711ec78a79663acb6e82.1456734966.git.leo@famulari.name>
From: Leo Famulari <leo@famulari.name>
Date: Mon, 29 Feb 2016 02:23:43 -0500
Subject: [PATCH 1/1] gnu: ilmbase: Add patches to fix build on i686.

Fixes <http://bugs.gnu.org/22049>.

* gnu/packages/patches/ilmbase-testBox.patch,
gnu/packages/patches/ilmbase-testBoxAlgo.patch: New files.
* gnu-system.am (dist_patch_DATA): Add them.
* gnu/packages/graphics.scm (ilmbase)[native-inputs]: Add patch/testBox
and patch/testBoxAlgo.
[arguments]: Add 'patch-for-i686' phase.
---
 gnu-system.am                                  |   2 +
 gnu/packages/graphics.scm                      |  29 ++++++
 gnu/packages/patches/ilmbase-testBox.patch     | 121 +++++++++++++++++++++++++
 gnu/packages/patches/ilmbase-testBoxAlgo.patch |  41 +++++++++
 4 files changed, 193 insertions(+)
 create mode 100644 gnu/packages/patches/ilmbase-testBox.patch
 create mode 100644 gnu/packages/patches/ilmbase-testBoxAlgo.patch

diff --git a/gnu-system.am b/gnu-system.am
index 502993b..42673c5 100644
--- a/gnu-system.am
+++ b/gnu-system.am
@@ -529,6 +529,8 @@ dist_patch_DATA =						\
   gnu/packages/patches/icu4c-CVE-2014-6585.patch		\
   gnu/packages/patches/icu4c-CVE-2015-1270.patch		\
   gnu/packages/patches/icu4c-CVE-2015-4760.patch		\
+  gnu/packages/patches/ilmbase-testBox.patch			\
+  gnu/packages/patches/ilmbase-testBoxAlgo.patch		\
   gnu/packages/patches/imagemagick-test-segv.patch		\
   gnu/packages/patches/irrlicht-mesa-10.patch			\
   gnu/packages/patches/jasper-CVE-2007-2721.patch		\
diff --git a/gnu/packages/graphics.scm b/gnu/packages/graphics.scm
index 560febe..0d99b4e 100644
--- a/gnu/packages/graphics.scm
+++ b/gnu/packages/graphics.scm
@@ -1,6 +1,7 @@
 ;;; GNU Guix --- Functional package management for GNU
 ;;; Copyright © 2015 Ludovic Courtès <ludo@gnu.org>
 ;;; Copyright © 2015 Tomáš Čech <sleep_walker@gnu.org>
+;;; Copyright © 2016 Leo Famulari <leo@famulari.name>
 ;;;
 ;;; This file is part of GNU Guix.
 ;;;
@@ -91,6 +92,34 @@ many more.")
                (base32
                 "1izddjwbh1grs8080vmaix72z469qy29wrvkphgmqmcm0sv1by7c"))))
     (build-system gnu-build-system)
+
+    ;; ilmbase fails to build from source on the i686 architecture due to a
+    ;; rounding issue. These patches are applied only when building on or for
+    ;; i686.
+
+    ;; Please refer to:
+    ;; https://lists.nongnu.org/archive/html/openexr-devel/2015-12/msg00001.html
+    ;; https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815712
+    (native-inputs
+     (if (string-prefix? "i686" (or (%current-target-system) (%current-system)))
+         `(("patch/testBox"
+            ,(search-patch "ilmbase-testBox.patch"))
+           ("patch/testBoxAlgo"
+            ,(search-patch "ilmbase-testBoxAlgo.patch")))
+         '()))
+    (arguments
+     (if (string-prefix? "i686" (or (%current-target-system) (%current-system)))
+         `(#:phases
+           (modify-phases %standard-phases
+             (add-after 'unpack 'patch-for-i686
+               (lambda* (#:key inputs #:allow-other-keys)
+                 (let ((patches (list (assoc-ref inputs "patch/testBox")
+                                      (assoc-ref inputs "patch/testBoxAlgo"))))
+                   (and (map (lambda (patches)
+                               (zero?
+                                 (system* "patch" "-p1" "--input" patches)))
+                              patches)))))))
+         '()))
     (home-page "http://www.openexr.com/")
     (synopsis "Utility C++ libraries for threads, maths, and exceptions")
     (description
diff --git a/gnu/packages/patches/ilmbase-testBox.patch b/gnu/packages/patches/ilmbase-testBox.patch
new file mode 100644
index 0000000..a947df3
--- /dev/null
+++ b/gnu/packages/patches/ilmbase-testBox.patch
@@ -0,0 +1,121 @@
+Fix FTBFS on i686-linux due to rounding issue (see references).
+
+Fixes Guix bug #22049 (see below).
+
+Copied from Debian.
+
+Source:
+https://sources.debian.net/src/ilmbase/2.2.0-11/debian/patches/testBoxAlgo.patch/
+
+References:
+https://lists.nongnu.org/archive/html/openexr-devel/2015-12/msg00001.html
+https://debbugs.gnu.org/cgi/bugreport.cgi?bug=22049
+https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815712
+https://anonscm.debian.org/cgit/pkg-phototools/ilmbase.git/commit/?id=ab28bb45cdad8adc32e345b777ab8e692b1d9a9c
+
+---
+
+Subject: testBox: allow fuzzy comparison of floats, doubles
+From: Steven Chamberlain <steven@pyro.eu.org>
+Date: Wed, 24 Feb 2016 01:10:11 +0000
+
+Allow for inexact values, as long as the error is smaller than the
+epsilon of the data type.
+
+On 32-bit x86, allow even greater discrepency at double
+precision, due to possible double-rounding.  See
+https://lists.nongnu.org/archive/html/openexr-devel/2015-12/msg00001.html
+
+Index: ilmbase/ImathTest/testBox.cpp
+===================================================================
+--- ilmbase.orig/ImathTest/testBox.cpp
++++ ilmbase/ImathTest/testBox.cpp
+@@ -47,6 +47,58 @@ using namespace IMATH_INTERNAL_NAMESPACE
+ 
+ namespace {
+ 
++template <class T>
++bool
++approximatelyEqual (const T &p1, const T &p2)
++{
++	/* int and short should be exact */
++	return (p1 == p2);
++}
++
++bool
++approximatelyEqual (const Vec2<float> &p1, const Vec2<float> &p2)
++{
++    float e = limits<float>::epsilon();
++    float m = 0;
++
++    for (int i = 0; i < 2; ++i)
++    {
++	m = max (m, abs (p1[i]));
++	m = max (m, abs (p2[i]));
++    }
++
++    for (int i = 0; i < 2; ++i)
++	if (!equalWithAbsError (p1[i], p2[i], m * e))
++	    return false;
++
++    return true;
++}
++
++bool
++approximatelyEqual (const Vec2<double> &p1, const Vec2<double> &p2)
++{
++#if defined(__i386__) || defined(_M_IX86)
++    /* double-rounding on 32-bit x86 may cause larger error:
++       use epsilon of float rather than double */
++    double e = limits<float>::epsilon();
++#else
++    double e = limits<double>::epsilon();
++#endif
++    double m = 0;
++
++    for (int i = 0; i < 2; ++i)
++    {
++	m = max (m, abs (p1[i]));
++	m = max (m, abs (p2[i]));
++    }
++
++    for (int i = 0; i < 2; ++i)
++	if (!equalWithAbsError (p1[i], p2[i], m * e))
++	    return false;
++
++    return true;
++}
++
+ //
+ // Test case generation utility - create a vector of IMATH_INTERNAL_NAMESPACE::Vec{2,3,4}
+ // with all permutations of integers 1..T::dimensions().
+@@ -250,7 +302,8 @@ testExtendByPoint(const char *type)
+                               
+         IMATH_INTERNAL_NAMESPACE::Box<T> b;
+         b.extendBy(p);
+-        assert(b.min == p && b.max == p);
++        assert (approximatelyEqual (b.min, p));
++        assert (approximatelyEqual (b.max, p));
+     }
+ 
+     //
+@@ -283,7 +336,8 @@ testExtendByPoint(const char *type)
+ 
+             b.extendBy(p);
+ 
+-            assert(b.min == min && b.max == max);
++            assert (approximatelyEqual (b.min, min));
++            assert (approximatelyEqual (b.max, max));
+         }
+     }
+ }
+@@ -358,7 +412,8 @@ testExtendByBox(const char *type)
+             }
+             b.extendBy(IMATH_INTERNAL_NAMESPACE::Box<T>(p0, p1));
+ 
+-            assert(b.min == min && b.max == max);
++            assert (approximatelyEqual (b.min, min));
++            assert (approximatelyEqual (b.max, max));
+         }
+     }
+ }
diff --git a/gnu/packages/patches/ilmbase-testBoxAlgo.patch b/gnu/packages/patches/ilmbase-testBoxAlgo.patch
new file mode 100644
index 0000000..06752e3
--- /dev/null
+++ b/gnu/packages/patches/ilmbase-testBoxAlgo.patch
@@ -0,0 +1,41 @@
+Fix FTBFS on i686-linux due to rounding issue (see references).
+
+Fixes Guix bug #22049 (see below).
+
+Copied from Debian.
+
+Source:
+https://sources.debian.net/src/ilmbase/2.2.0-11/debian/patches/testBox.patch/
+
+References:
+https://lists.nongnu.org/archive/html/openexr-devel/2015-12/msg00001.html
+https://debbugs.gnu.org/cgi/bugreport.cgi?bug=22049
+https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815712
+https://anonscm.debian.org/cgit/pkg-phototools/ilmbase.git/commit/?id=ab28bb45cdad8adc32e345b777ab8e692b1d9a9c
+
+---
+
+Subject: testBoxAlgo: allow fuzzy match of b12 == b2
+From: Steven Chamberlain <steven@pyro.eu.org>
+Date: Wed, 24 Feb 2016 01:04:11 +0000
+
+Also fix a pre-existing typo.
+
+Index: ilmbase/ImathTest/testBoxAlgo.cpp
+===================================================================
+--- ilmbase.orig/ImathTest/testBoxAlgo.cpp
++++ ilmbase/ImathTest/testBoxAlgo.cpp
+@@ -886,10 +886,11 @@ boxMatrixTransform ()
+ 
+     assert (approximatelyEqual (b2.min, b4.min, e));
+     assert (approximatelyEqual (b2.max, b4.max, e));
+-    assert (approximatelyEqual (b3.max, b4.max, e));
++    assert (approximatelyEqual (b3.min, b4.min, e));
+     assert (approximatelyEqual (b3.max, b4.max, e));
+ 
+-    assert (b21 == b2);
++    assert (approximatelyEqual (b2.min, b21.min, e));
++    assert (approximatelyEqual (b2.max, b21.max, e));
+     assert (b31 == b3);
+ 
+     M[0][3] = 1;
-- 
2.7.1