Control: severity -1 serious
On Thu, Jan 25, 2018 at 09:53:21PM +0200, Adrian Bunk wrote:
Control: tags -1 - moreinfo
On Thu, Jan 25, 2018 at 07:39:27PM +0100, Julien Cristau wrote:
Control: severity -1 normal
Control: tag -1 moreinfo
On Thu, Jan 25, 2018 at 14:11:49 +0200, Adrian Bunk wrote:
Package: libmpfr6
Version: 4.0.0-5
Severity: serious
Mixing libmpfr4 and libmpfr6 doesn't work well:
flint-arb FTBFS with:
/usr/bin/ld: warning: libmpfr.so.4, needed by /usr/lib/libflint.so, may conflict with libmpfr.so.6
/usr/bin/ld: mpfr_free_func: TLS definition in //usr/lib/x86_64-linux-gnu/libmpfr.so.4 section .tbss mismatches non-TLS definition in /usr/lib/gcc/x86_64-linux-gnu/7/../../../x86_64-linux-gnu/libmpfr.so section .text
//usr/lib/x86_64-linux-gnu/libmpfr.so.4: error adding symbols: Bad value
collect2: error: ld returned 1 exit status
Some packages like fractalnow FTBFS when gcc and libmpc3 use
different mpfr libraries, with a gcc ICE:
../../src/init2.c:52: MPFR assertion failed: p >= 2 && p <= ((mpfr_prec_t)((mpfr_uprec_t)(~(mpfr_uprec_t)0)>>1))
src/fractal_compute_engine.c: In function 'FractalLoopMANDELBROTPINTAVERAGECOLORINGDISCRETECURVATURENONESINGLE':
src/fractal_compute_engine.c:285:1: internal compiler error: Aborted
It is not even obvious in the latter case that this is always only an ICE,
and not sometimes a miscompilation.
The libmpc3 issue is also expected to hit users who have older gcc versions
still installed, e.g. gcc-6 still installed after stretch->buster upgrade.
When the dependencies are fulfilled users can expect to have working software,
even a forced removal on stretch->buster upgrades is better than runtime problems.
Is this actually a problem between libmpfr4 and libmpfr6, or libmpfr4
and the new libmpfr-dev?
...
This is a problem between libmpfr4 and libmpfr6.
libmpfr-dev is not installed when gcc ICEs building mathgl.[1,2]
...
It is not even obvious that we will always be lucky and get an ICE,
it is even possible that in some cases gcc might end up silenty
miscompiling code.
ICE or worse, this would be a problem for anyone still having gcc-6
or older compiler packages from previous releases installed when
upgrading to buster.
And without an RC bug it would already cause problems much earlier for
derivates based on testing, since mpfr4 and mpclib3 would currently
migrate to testing before gcc-7 migrates.
cu
Adrian