Re: Revised patch for Ada bootstrap

[lutz at iks-jena dot de (Lutz Donnerhacke)]

* Zack Weinberg wrote:
>and having generated the missing source files by hand, I can get all
>the way through a 3-stage bootstrap.  I then get comparison failures
>for all object files whose source is Ada.

Applying

diff -r1.748 Makefile.in
3108a3109,3111
> # Continue bootstrap on known failtures:
> # sed 's-\(.*\) differs-stage2/\1-' gcc/.bad_compare >
gcc/.known_bad_compare 
> 
3113a3117
>         grep -q ^stage$$stage/$$file$$ .known_bad_compare && continue; \
3121a3126
>             grep -q ^stage$$stage/$$file$$ .known_bad_compare && continue; \
3148a3154
>         grep -q ^stage$$stage/$$file$$ .known_bad_compare && continue; \
3154a3161
>             grep -q ^stage$$stage/$$file$$ .known_bad_compare && continue; \


allows to go further and results in a 'gnatbind.ali Format error'.
Reproducing this (seem a PATH problem using the 3.13p suite) was not possible:
  make -C gcc restage1
followed by
  make bootstrap
runs without problem.

>In addition to the obvious ordering changes and associated changes in
>type numbering, many but not all of the stabs have their DESC field
>(second from the right) changed from 19 to 1.  I have not been able to
>find any documentation of what that field means for type stabs.

According to http://www.math.utah.edu/docs/info/stabs_12.html is is a line
number which can be changed due to the already noticed reposition action.
It seems to be only used by the debugger, but the DESC is not really useful
for them. gcc/dbxout.c explains the whole format very clearly and seems to
require DBX_USE_BINCL to fill the DESC field with a 'file_number' (not sure).