Re: [PATCH][gdb/testsuite] Fix error message test in dw2-error.exp

[Pedro Alves <palves@redhat.com>]

On 06/19/2018 10:00 AM, Tom de Vries wrote:

> the executable used in dw2-error.exp is compiled from a .s that was generated
> with dwarf2 debug information but has been hand-edited to set the version in
> the compilation unit header to 0x99:
> ...
> .Ldebug_info0:
>         .long   0x4e    # Length of Compilation Unit Info
>         .value  0x99    # DWARF version number
>         .long   .Ldebug_abbrev0 # Offset Into Abbrev. Section
> ...
> 
> Consequently, dwarf2read.c:read_comp_unit_head() interprets the compilation
> unit header as dwarf5, 


That right there looks like the real bug to me.

I went looking for the history behind the testcase, and
got surprised that the testcase is expecting that "wrong unit_type in
compilation unit header" error instead of the same error that had been
reported in the original bug report at
<https://sourceware.org/bugzilla/show_bug.cgi?id=14931>:

 ~~~~~
 Dwarf Error: wrong version in compilation unit header (is 4, should be 2) [in module ....build/gdb/gdb]
 ~~~~~

read_and_check_comp_unit_head calls error_check_comp_unit_head 
after calling read_comp_unit_head, and thus AFAICT error_check_comp_unit_head
would error out with the "wrong version" error, the one that had been
reported in the original bug report.  That seems like a much better
error to me.

static void
error_check_comp_unit_head (struct dwarf2_per_objfile *dwarf2_per_objfile,
			    struct comp_unit_head *header,
			    struct dwarf2_section_info *section,
			    struct dwarf2_section_info *abbrev_section)
{
  const char *filename = get_section_file_name (section);

  if (header->version < 2 || header->version > 5)
    error (_("Dwarf Error: wrong version in compilation unit header "
	   "(is %d, should be 2, 3, 4 or 5) [in module %s]"), header->version,
	   filename);



So it seems to me that read_comp_unit_head shouldn't be
trying to interpret contents of a dwarf version that 
gdb doesn't understand.  Seems like that error_check_comp_unit_head 
version check is too late?  How about moving it into
read_and_check_comp_unit_head?  Of course, the testcase would then
be adjusted to expect the new message, and it would expect 153/0x99
exactly instead of any number, which ensures that gdb reads and
prints the version number correctly.

Thanks,
Pedro Alves

> which starts with fields unit_length (4 or 12 byte
> unsigned), version (uhalf), and unit_type (ubyte). So, the unit_type
> field is initialized from the first byte of .Ldebug_abbrev0 offset.
> 
> Using objdump, we find that the value of that byte is 0x64.
> ...
> Contents of section .debug_info:
>  ...
>  00c0 00450000 0001804e 00000099 00640000  .E.....N.....d..
> ...
> 
> And indeed gdb errors out accordingly (note: 0x64 == 100):
> ...
> (gdb) file outputs/gdb.dwarf2/dw2-error/dw2-error
> Reading symbols from outputs/gdb.dwarf2/dw2-error/dw2-error...
> Dwarf Error: wrong unit_type in compilation unit header
>              (is 100, should be 1 or 2)
> 	     [in module outputs/gdb.dwarf2/dw2-error/dw2-error]
> (no debugging symbols found)...done.
> ...
> 
> The test fails however because it expects the error message to contain 0
> instead of the 100 we're seeing.
> 
> This patch fixes the failure by allowing any value for the unit_type in the
> error message.