[Pkg-clamav-devel] tomfastmath is ready

[Andreas Cadhalpun]

Hi Sebastian,

On 10.03.2015 09:07, Sebastian Andrzej Siewior wrote:
> On 2015-02-06 20:56:24 [+0100], Andreas Cadhalpun wrote:
>> There are also some other tomfastmath changes in clamav.
>> In the upstream git repo one finds:
> finally I'm done with it.

Great!

>> commit c424c9301376091282cf64df7eae8629682c140a
>> Author: David Raynor <draynor at sourcefire.com>
>> Date:   Wed Feb 13 15:35:01 2013 -0500
>>
>>      cid #11125/#11126 fixed
>>
>> commit 627e8835f113cbad18afc0773c4b6e91f5826f20
>> Author: David Raynor <draynor at sourcefire.com>
>> Date:   Wed Feb 13 15:10:41 2013 -0500
>>
>>      cid #11125/#11126
>
> those two are not bug ids but someting from an auto-testing tool. It
> does not make sense.

These are coverity IDs, but it's not obvious what they are about.
If upstream also doesn't understand it, it should be fine to ignore them.

>> Can you forward these fixes to tomsfastmath upstream as well?
>
> I forwarded what made sense upstream. I backported around 20 patches
> from upstream. It compiles without warnings on amd64 and that one
> warning in fp_ident on i386 is minor. I will look at it later.
> There is one patch which I need mail upstream. I compiled clamav on i386
> against it and the testsuite passes. Haven't look at the build log yet.

OK.

> I defined four symbols optional which are only exported on 32bit
> architectures. The longterm plan to fix this properly is to wait for
> dpkg 1.18 which defines the tag arch-bits= or to ask upstream those
> symbols are worth exporting or should remain for internal usage. I tend
> to the later.

As long as nothing uses these symbols, it doesn't matter much.

I pushed a patch, which allows building clamav against the system 
tomsfastmath library, if it is detected and otherwise uses the embedded 
copy.
With the current libtfm it seems to work fine, so once it is in the 
archive, we can just replace the (useless) libtommath-dev dependency 
with libtfm-dev.

Best regards,
Andreas