Fortran Common Blocks

[davea at quasar dot engr dot sgi dot com (David B Anderson)]

"Knott, Martin" <martin@epc.co.uk> writes:
>I am currently attempting to generate Dwarf2 information for common blocks.
>However, I am a little curious by the messages that DwarfDump produces:
>
><1><  123>      DW_TAG_subprogram
>                DW_AT_decl_line             1
>                DW_AT_decl_column           0
>                DW_AT_decl_file             1
>                DW_AT_sibling               <192>
>                DW_AT_accessibility         DW_ACCESS_public
>                DW_AT_name                  main.program
>                DW_AT_type                  <115>
>>                DW_AT_prototyped            no
>                DW_AT_high_pc               0x8048d20
>>                DW_AT_low_pc                0x8048c50
>                DW_AT_external              yes(1)
>*** DWARF CHECK: DW_TAG_common_block: tag-tree relation is not valid ***
><2><  159>      DW_TAG_common_block
>                DW_AT_decl_line             6


As James Cownie writes, this is sensible dwarf.

You will find the input that causes  the 
validity table to be built In the dwarfdump source, in the file
tag_tree.list under DW_TAG_subprogram there is no
DW_TAG_common_block. That is a bug in the table.

Note the table uses 2 32bit ints for bit flags rather than a single
64bit int : when dwarfdump was written, 'long long' was
not considered quite 'normal'... 

SGI f77 generates for example, on James's example:
<1><  128>      DW_TAG_subprogram
                DW_AT_decl_file             1
                DW_AT_decl_line             2
                DW_AT_decl_column           18
                DW_AT_name                  foo
                DW_AT_external              yes(1)
                DW_AT_frame_base            DW_OP_breg29+0
                DW_AT_low_pc                0
                DW_AT_high_pc               0x8
                DW_AT_MIPS_fde              32
                DW_AT_sibling               <209>
<2><  156>      DW_TAG_common_inclusion
                DW_AT_decl_line             5
                DW_AT_decl_column           15
                DW_AT_common_reference      <163>
<2><  163>      DW_TAG_common_block
                DW_AT_name                  madman
                DW_AT_location              DW_OP_addr 0
<3><  177>      DW_TAG_variable
 ...

and gets the same warning with -k.

Seems like making the checking be optional was not such a good idea,
as I never think to turn it on.

Anyway, fix your copy of the table.
Also, add
DW_TAG_variable under
the
DW_TAG_common_block entry
in the list.
Likely there are other botches in the list too.
Let me know of those you find.

I'll fix the master source.

It's time for a re-distribution of dwarfdump, but that may not
be for a few days yet.



davea@sgi.com