Bugzilla Bug 152818

Please test hspell and re-keyword kdelibs (together with hspell) asap. Thanks.

Alon, we might need a version of hspell-1.0 that builds a .so, as it needs to
use PIC mode for AMD64 most likely.

Created an attachment (id=100484) [edit]
hspell-1.0-shared.patch

This patch adds support for libhspell.so, that will in turn allow kdelibs to
build correctly against hspell on arches where PIC is needed for shared
libraries (AMD64 and IA64 comes to mind).

AMD64 and the rest of PIC-based arches, please wait till this is merged in -r1
for keywording, or kdelibs will be broken again.

Diego: The hspell2 patch doesn't solve the problem that aspell and ispell are
automagically detected, when building the kspell2 plugins... 

Uh, aspell is not automagically detected (--without-aspell works for me).
ispell maybe, let me check twice, but it should be detected at _runtime_ which
is a good kind of automagic dependency.

Checked twice: ispell support is *always* built, and ispell is detected at
runtime if present, so there's no need to support it as useflag in kdelibs
(really, *who* would be using ispell?).

Aspell in KSpell2 will be enabled/disabled by the spell useflag, that by
coincidence will also bring it in for the KSpell1 support (that is discovered
automatically at runtime).

HSpell in KSpell2 will be enabled/disabled by linguas_he plus spell useflag,
that by coincidence will also bring it in for the KSpell1 support (that is
discovered automatically at runtime).

Which means that as it is, it works, no automagic linking dependencies.

Forget about the automagic dependencies popping into my mind seeing you
readding kspell2 support. What I really had vaguely remembered was the former
existing spell? || dependency issue, but since you force aspell on hebrew
speaking users, this isn't an issue.

aspell is used for everything else but Hebrew, so it's just simpler to force it
on for everybody, this way we solve the problem for good. Maybe users who only
speak Hebrew and don't care about other languages' spellcheckers might want to
get rid of aspell, but as that requires an higher complexity solution, I'd say
we're good as we are :)

(PPC64)

hspell does not like parallel building (MAKEOPTS="-j4"):

[...]
rm -f hebrew.wgz.lingsizes.tmp
rm -f hebrew.wgz.prefixes
gzip -9 < hebrew.wgz.prefixes.tmp >hebrew.wgz.prefixes
/bin/sh: hebrew.wgz.prefixes.tmp: No such file or directory
make: *** [hebrew.wgz.stems] Error 1

Created an attachment (id=100512) [edit]
hspell-1.0-shared.patch

This patch works with --as-needed and fixes libhspell.a building too.
Now building kdelibs to see if it works as expected.

Thanks for all the help!
I will look at it all today.

Regarding the aspell issue, there is no problem in forcing aspell for all
users.

Created an attachment (id=100517) [edit]
Just a simple unicode text in hebrew with few errors for testing hspell

Diego,

First, I got similar failure while installing on x86 when MAKEOPTS more than 1,
just got it on different file.

Second, I attached some simple paragraph for testing the hspell in the future
upon your request. It has few errors in it.

If there's any questions about it, just let me know.

Created an attachment (id=100518) [edit]
a newer simple file

Previous file got corrupted

Thanks Alex :)

For the ones testing, if your locale is UTF-8, just use "iconv -f UTF-16 file"
to get the transcoded version.

To run the hspell utility by hand, I've used the command "iconv -f UTF-16 -t
ISO-8859-8 hebrew | hspell -c | iconv -f ISO-8859-8".. I suppose it works as it
outputs something, but I cannot read it so I just hope it's correct ;)

I've just confirmed that hspell kspell2 plugin builds (and seems to work) fine
with the patch.

(In reply to comment #8)
> (PPC64)
> 
> hspell does not like parallel building (MAKEOPTS="-j4"):
> 
> [...]
> rm -f hebrew.wgz.lingsizes.tmp
> rm -f hebrew.wgz.prefixes
> gzip -9 < hebrew.wgz.prefixes.tmp >hebrew.wgz.prefixes
> /bin/sh: hebrew.wgz.prefixes.tmp: No such file or directory
> make: *** [hebrew.wgz.stems] Error 1
> 

Diego,

1. The same error with -j4 I have on AMD64. Sometime it installs, sometimes it
doesn't.
2. With MAKEOPTS="-j1" install is successful in 100%
    Checked with the following:
         set temp PORT_LOGDIR location in /etc/make.conf
        run the following: for i in 1 2 3 4 5 6 7 8 9 10; do emerge hspell;
done
        checked a log - all success.
        removed temp PORT_LOGDIR from /etc/make.conf
3. Given example file is successfully tested. I see some misses, but its hspell
issue, not ebuild.

Anything else I can help with?

forgot to mention, all above is regarding hspell-1.0

Use -j1... I will fix this in my next commit.

I've committed somewhat larger (a lot) patch, to work with automake, continue
of work started at upstream bugzilla. I hope upstream will merge this one as it
work correctly.

It also solves the parallel make issue, pkg-config issue and more.

Please check it out.

(In reply to comment #18)

AMD64, hspell-1.0-r1

Compiles fine with -j4.
Works fine.

I have updated bug#152274, in order to add all kdelibs lost keywords into
hspell.

~amd64 re-added.

~x86-fbsd re-added.

added ~ppc64

~sparc is back!

This one is stable on hppa. Now dealing with -r6.

Going to unify with bug #155505.

*** Bug 155505 has been marked as a duplicate of this bug. ***

SPARC done

HPPA done.

-r10 stable on mips.