[Issue 3956] linker removes underscore from all exported symbols of a module but the first

[d-bugmail at puremagic.com]

http://d.puremagic.com/issues/show_bug.cgi?id=3956



--- Comment #5 from Martin Nowak <code at dawg.eu> 2013-09-28 07:45:37 PDT ---
YES_EXPORT PROC
;; ...
#ifdef fg_pe
   if (OUTPUT_PE) // false on first test
      TWEAK_EXPORT_NAMES();
#endif
    DEFINE_EXPORT(); // could set ANY_USE32, but it's usually already
    SELECT_OUTPUT_SEGMENTED_OR_PE();
;; ...
YES_EXPORT ENDP

SELECT_OUTPUT_SEGMENTED_OR_PE PROC
    if (!(OUTPUT_SEGMENTED | OUTPUT_PE))
       return;
    if (ANY_USE32) // this is already set by defining the output segments
        SELECT_OUTPUT_PE(); // sets OUTPUT_PE
    else
        SELECT_OUTPUT_SEGMENTED();
SELECT_OUTPUT_SEGMENTED_OR_PE ENDP

I found the bug, thanks to OllyDbg, but I could use some help to interpret and
resolve it.
>From what I understand optlink determines the output format OUTPUT_SEGMENTED or
OUTPUT_PE depending on the input files. That is when it first encounters some
usage of 32-bit symbols it will set ANY_USE32.
Striping the leading underscore in the exported name is only done for
OUTPUT_PE.
The problem comes from the circular dependency in the OUTPUT_PE test above. For
the first export symbol no output format has yet been set, thus the test fails
and the symbol is not stripped.

One way to resolve this would be to prematurely test for ANY_USE32 if no OUTPUT
is selected.

    if (OUTPUT_PE || !OUTPUT_SEGMENTED && ANY_USE32)


Any ideas?

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------