Re: [PATCH] convert braced initializers to strings (PR 71625)

[Martin Sebor <msebor at gmail dot com>]

On 08/15/2018 03:34 PM, Jeff Law wrote:
> On 08/15/2018 03:02 PM, Martin Sebor wrote:
> > On 08/15/2018 06:07 AM, Joseph Myers wrote:
> > > On Tue, 14 Aug 2018, Martin Sebor wrote:
> > >
> > > > > This is with Bison 3.0.4, should the version used to produce
> > > > > intl/plural.c
> > > > > prove relevant.
> > > >
> > > > Can you send me the translation unit and the options it was compiled
> > > > with that triggered the errors?
> > >
> > > I've attached plural.i.  The error is a static link error linking sln,
> > > but
> > > maybe comparing results of compiling plural.i before and after the
> > > changes
> > > may be enlightening (unless it's actually a difference in code elsewhere
> > > in glibc causing a link error reported in plural.o).
> > >
> > > Compiled with:
> > >
> > > alpha-glibc-linux-gnu-gcc
> > > /scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu/intl/plural.c
> > >
> > > -c -std=gnu11 -fgnu89-inline  -O2 -Wall -Werror -Wundef -Wwrite-strings
> > > -fmerge-all-constants -fno-stack-protector -frounding-math -g
> > > -Wstrict-prototypes -Wold-style-definition -fno-math-errno
> > > -mlong-double-128 -mieee -mfp-rounding-mode=d
> > > -ftls-model=initial-exec
> > > -I../include
> > > -I/scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu/intl
> > >
> > > -I/scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu
> > >
> > > -I../sysdeps/unix/sysv/linux/alpha/alpha
> > > -I../sysdeps/unix/sysv/linux/alpha/fpu  -I../sysdeps/alpha/fpu
> > > -I../sysdeps/unix/sysv/linux/alpha  -I../sysdeps/alpha/nptl
> > > -I../sysdeps/unix/sysv/linux/wordsize-64
> > > -I../sysdeps/ieee754/ldbl-64-128
> > > -I../sysdeps/ieee754/ldbl-opt  -I../sysdeps/unix/sysv/linux/include
> > > -I../sysdeps/unix/sysv/linux  -I../sysdeps/nptl  -I../sysdeps/pthread
> > > -I../sysdeps/gnu  -I../sysdeps/unix/inet  -I../sysdeps/unix/sysv
> > > -I../sysdeps/unix/alpha  -I../sysdeps/unix  -I../sysdeps/posix
> > > -I../sysdeps/alpha  -I../sysdeps/wordsize-64
> > > -I../sysdeps/ieee754/ldbl-128  -I../sysdeps/ieee754/dbl-64/wordsize-64
> > > -I../sysdeps/ieee754/dbl-64  -I../sysdeps/ieee754/flt-32
> > > -I../sysdeps/ieee754  -I../sysdeps/generic  -I.. -I../libio -I.
> > > -D_LIBC_REENTRANT -include
> > > /scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu/libc-modules.h
> > >
> > > -DMODULE_NAME=libc -include ../include/libc-symbols.h
> > > -DTOP_NAMESPACE=glibc -D'LOCALEDIR="/usr/share/locale"'
> > > -D'LOCALE_ALIAS_PATH="/usr/share/locale"' -o
> > > /scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu/intl/plural.o
> > >
> > > -MD -MP -MF
> > > /scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu/intl/plural.o.dt
> > >
> > > -MT
> > > /scratch/jmyers/glibc/many9/build/compilers/alpha-linux-gnu/glibc/alpha-linux-gnu/intl/plural.o
> >
> > Thanks.  I don't see anything obviously wrong but I don't know
> > much about Alpha assembly.  Attached are the two .s files, with
> > (plural-new.s) and without (plural-old.s) the array-to-string
> > transformation.
> I'd focus on these  insns which correspond to the error from the linker:
>
> They're in plural.o within the fucntion gettextparse
>
>
>         ldah $2,yypgoto+2305843009213693936($29)                !gprelhigh
>
> Something certainly doesn't look right...

Thanks.  I also get the same exact same assembly with both
forms of initializers for the two constant arrays, i.e.,

  static const yytype_int8 yypgoto[3] = "\366\366\377";
  static const yytype_int8 yydefgoto[3] = "\377\005\006";

and

  static const yytype_int8 yypgoto[] = { -10, -10, -1 };
  static const yytype_int8 yydefgoto[] = { -1, 5, 6 };

They end up in the .sdata section either way (the former
both with and without the GCC change as might be expected):

        .section        .sdata,"aws"
        .type   yydefgoto, @object
        .size   yydefgoto, 3
  yydefgoto:
        .ascii  "\377\005\006"
        .type   yypgoto, @object
        .size   yypgoto, 3
  yypgoto:
        .ascii  "\366\366\377"
        .section        .rodata

and also before the GCC change:

        .section        .sdata,"aws"
        .type   yydefgoto, @object
        .size   yydefgoto, 3
  yydefgoto:
        .byte   -1
        .byte   5
        .byte   6
        .type   yypgoto, @object
        .size   yypgoto, 3
  yypgoto:
        .byte   -10
        .byte   -10
        .byte   -1

> At the .o level this turns into:
> 218:   00 00 5d 24     ldah    t1,0(gp)
>                         218: GPRELHIGH  .sdata+0x1ffffffffffffff3
>
>
> 300:   00 00 5d 24     ldah    t1,0(gp)
>                         300: GPRELHIGH  .sdata+0x1ffffffffffffff3
>
> It's not really a matter of what the instruction does, but a matter of
> the relocation.  You'd have to look at the definition of GPRELHIGH which
> you can find in BFD.

The definitions match the assembly with both kinds of
initializers, and with the string literal also with GCC before
the change:

  0000000000000218 GPRELHIGH         .sdata+0x1ffffffffffffff3
  0000000000000238 GPRELLOW          .sdata+0x1ffffffffffffff3
  0000000000000300 GPRELHIGH         .sdata+0x1ffffffffffffff3
  0000000000000308 GPRELLOW          .sdata+0x1ffffffffffffff3
  0000000000000458 GPRELHIGH         .sdata+0x0000000000000003
  0000000000000468 GPRELLOW          .sdata+0x0000000000000003

vs the array form before the GCC change:

  0000000000000468 GPRELHIGH         .sdata+0x0000000000000003
  0000000000000488 GPRELLOW          .sdata+0x0000000000000003
  000000000000057c GPRELHIGH         .sdata
  0000000000000580 GPRELLOW          .sdata

So it seems as though using the string literal as an initializer
tickles a latent bug in GCC and the question is where.

Martin