libreoffice ftbfs on armhf

The ARM Procedure Call Standard document on http://infocenter.arm.com/ has details of the VFP variant of the procedure call standard, required for armhf.

Simplified, each floating-point argument is allocated first to the VFP registers if there are some available (from s0-s15 / d0-d7), and then to the stack when there are no more VFP argument registers available.

Basically, for the armhf case we need:

  * A separate array to track the values to be passed to the called function in floating-point registers
  * A separate counter to track which floating-point argument registers have already been allocated
  * Separate implementations of INSERT_FLOAT() and INSERT_DOUBLE() to allocate to VFP regs and then to the stack (similarly to the way the INSERT_INT*() macros work)
  * Code to copy the required arguments into the VFP regs when the actual target function is called
  * Code to handle floating-point return values (float/double returned VFP registers) -- refer to the Procedure Call Standard for exact details)

http://cgit.freedesktop.org/libreoffice/core/tree/bridges/source/cpp_uno/gcc3_linux_arm/uno2cpp.cxx

incomplete first attempt attached, including packaging updates.

Thanks Matthias! Bjoern, can you please take it from here? scheat.canonical.com has a precise-armhf dchroot for porting.

This is going to be a non-trivial porting project due to libreoffice needing to know a massive *ton* about how the C++ calling conventions on a given platform work.

Here's a general idea of what it took to build libreoffice (partially) on Windows x64 which explains the broken module: http://libreofficeaustralia.org/pa/community/blog/planet/20110328/porting-libreoffice-x64-windows

I'm reading through the ARM calling convention guide and using libffi's source as a handy reference guide, and will take a stab at doing the port, but its absolutely suck and pain.

@mcasadevall: If you need help for the LibreOffice side, please join #libreoffice-dev on freenode. sberg and caolan might be esp. helpful for low-level bridges stuff.

@bjoern-michaelsen: I am looking at the bridge code and will ask around the libo channel once I have proper questions.

Patch against git master, WIP. It still crashes in the bridge tests but it crashes a bit later.
Not yet for upstream commiting, still debugging it.

Updated patch agains master, parts of the previous one were commited in upstream git.

latest iteration, passes bridge tests in libo master tree.
subject to cleanups and further testing though

Great!
@jani: Could you check if the data at https://wiki.documentfoundation.org/Development/Developers is correct?

@jani: does this include the fix for softp as used by debian?

Diff agaist git master as of 3 days ago so it contains all 3 changesets, should be appliable to the Ubuntu/Debian package.

Three commits in git master fix this

This bug was fixed in the package libreoffice - 1:3.5.0-1ubuntu1

---------------
libreoffice (1:3.5.0-1ubuntu1) precise; urgency=low

  * Ubuntufy splash progress
  * merge from debian-experimental-3.5

libreoffice (1:3.5.0-2) UNRELEASED; urgency=low

  [ Rene Engelhard ]
  * debian/patches/armhf-bridges*.diff: add armhf bridges fixes from
    Jani Monoses (LP: #900636)
  * debian/patches/revert-468fe685e3c58c84bce6d9a48b931dcc21682679.diff:
    remove, obsolete
  * debian/patches/icu-arm.diff: fix internal icu build on arm.

  * debian/rules:
    - remove obsolete (and buggy) conditional icu-on-arm-fails breaking
      the build

  [ Bjoern Michaelsen ]
  * debian/libreoffice-common.postrm.in: remove mismerged superfluous fi

libreoffice (1:3.5.0-1ubuntu1~ppa2) precise; urgency=low

  * remove failing smoketest cleanup code for now

libreoffice (1:3.5.0-1ubuntu1~ppa1) precise; urgency=low

  * reenable EVO2 and MEDIAWIKI extensions (reduces delta to debian)
  * move unopkg wrapper from -common to -core to allow extension
    registration
  * do extension registration from -common to -core because we actually
    have the binaries (unopkg.bin etc.) to do it there (LP: #917153)
  * cherry-picked from debian-experimental-3.5:
    debian/libreoffice-common.postrm.in: remove mismerged superfluous fi
  * use source tarballs from debian

libreoffice (1:3.5.0-1) experimental; urgency=low

  * LibreOffice 3.5.0 final release (identical to rc3)

  * debian/patches/no-rcX-in-final.diff: don't show rc3, it's not a rc
    amymore...
  * debian/patches/patches/make-gengal-work-again.diff: register gengal.bin
    executable to get proper rpath and remove gengalrc
  * debian/patches/odk-fix-rdb-paths.diff: backport fix to fix rdb path
    in settings/std.mk in the SDK

  * debian/rules:
    - install found .gdb-py files into respective -dbg packages
    - add additional libpoppler-private-dev build-dep for new popplers
  * debian/control.in, debian/libreofice-common.p*:
    apply patch from Colin Watson to use Pre-Depends rather than
    'dpkg-maintscript-helper supports' guards (closes: #659867)
  * debian/watch: point to download.documentfoundation.org/libreoffice/src
    instead of dev-builds

libreoffice (1:3.5.0~rc3-1) experimental; urgency=low

  * new upstream release candidate

  * debian/patches/SwTblBoxFormula-destructor.diff: work around link error
    on armel by specifying explicit SwTblBoxFormula destructor, thanks
    Michael Meeks
  * debian/patches/ia64-oox-size.diff: build oox on ia64 with -Os
    as otherwise it gets too big to link
  * debian/patches/mediwiki-oor-replace.diff: replace oor:fuse by oor:replace
    for -wiki-publishers config (hopefully closes: #616011)

  * debian/rules:
    - give gb_FULLDEPS= to actual make check call, not debian/rules
      check
    - simplify -dbg/--enable-symbols conditionals
    - build with default boost again but add build-dep on g++-4.6 (>= 4.6.2-12)

libreoffice (1:3.5.0~rc2-2) experimental; urgency=low

  * debian/rules:
    - fix another merge error to really fix patch rule in check
    - enable ca-XV, put into -{l10n,help}-ca

libreoffice (1:3.5.0~rc2-1) experimental; urgency=low

  * new up...

FTBFS again much later at the install step.

The armhf build may be due to Base being enabled there (it is not on armel)
According to the Debian maintainer this should work around it

http://anonscm.debian.org/gitweb/?p=pkg-openoffice/libreoffice.git;a=blobdiff;f=rules;h=231d8a4724e6dc9e3a5572956c884ff54fe656b9;hp=d6ecdb2e3b321227128279c50165932ae6d46606;hb=17df62d42032133099da5a6533af0867335d1d8c;hpb=6bb9680004e13ba50cdc9a1d863fbc821a1e3e55

finally succeeded after disabling base on armhf too.