Re: [PATCH] Add "set debug minsyms" command

[John Baldwin <jhb at FreeBSD dot org>]

On 12/21/18 1:47 PM, Simon Marchi wrote:
> While discussing this issue:
> 
>   https://sourceware.org/ml/gdb-patches/2018-12/threads.html#00082
> 
> I added a printf to be able to quickly see all minimal symbols recorded
> by GDB.  I thought it would be useful to have it built-in, for the
> future.
> 
> The output isn't particularly pretty.  I found it more readable when making
> sure the fields were vertically aligned, which results in a lot of space wasted
> in the "type" column (the width is based on the length of the longest
> enumerator):
> 
>   minsym: recording minsym type: mst_data               addr: 0x00000000004047c0  section: 2      name: __rt_psrelocs_end
>   minsym: recording minsym type: mst_text               addr: 0x0000000000402b88  section: 0      name: exit
> 
> But since this is just debugging output, I think it doesn't really
> matter.
> 
> Also, I didn't use paddress to print the address, because:
> 
> 1. There is no gdbarch handy at this point
> 2. The address may not actually be an address, but any numerical value.
>    Printing with paddress could change how it's displayed (e.g. mask
>    certain bits) and could be misleading.  I think it's better to print
>    the actual raw value saved in the minimal symbol.

2) seems compelling to me.
 
> gdb/ChangeLog:
> 
> 	* minsyms.c: Include cli/cli-cmds.h.
> 	(debug_minsyms): New.
> 	(mst_str): New.
> 	(minimal_symbol_reader::record_full): Add debug output.
> 	(_initialize_minsyms): New.
> ---
>  gdb/auto-load.c |  2 ++
>  gdb/elfread.c   |  2 ++
>  gdb/minsyms.c   | 48 ++++++++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 52 insertions(+)
> 
> diff --git a/gdb/auto-load.c b/gdb/auto-load.c
> index 33d282afe83..e35fd29426b 100644
> --- a/gdb/auto-load.c
> +++ b/gdb/auto-load.c
> @@ -1175,6 +1175,8 @@ load_auto_scripts_for_objfile (struct objfile *objfile)
>  static void
>  auto_load_new_objfile (struct objfile *objfile)
>  {
> +  std::vector <int> c;
> +  c.clear();
>    if (!objfile)
>      {
>        /* OBJFILE is NULL when loading a new "main" symbol-file.  */

This seems spurious (also not in ChangeLog)?

> diff --git a/gdb/elfread.c b/gdb/elfread.c
> index 71e6fcca6ec..359089b166c 100644
> --- a/gdb/elfread.c
> +++ b/gdb/elfread.c
> @@ -249,6 +249,8 @@ elf_symtab_read (minimal_symbol_reader &reader,
>  	  continue;
>  	}
>  
> +      printf(" --- %s\n", sym->name);
> +
>        /* Skip "special" symbols, e.g. ARM mapping symbols.  These are
>  	 symbols which do not correspond to objects in the symbol table,
>  	 but have some other target-specific meaning.  */

Likewise.

> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
> index 0f854422e0f..98ce969eed0 100644
> --- a/gdb/minsyms.c
> +++ b/gdb/minsyms.c
> @@ -1112,6 +1143,11 @@ minimal_symbol_reader::record_full (const char *name, int name_len,
>    if (ms_type == mst_file_text && startswith (name, "__gnu_compiled"))
>      return (NULL);
>  
> +  if (debug_minsyms)
> +    printf_unfiltered
> +      ("minsym: recording minsym type: %-21s  addr: 0x%016llx  section: %-5d  name: %s\n",
> +       mst_str (ms_type), (long long) address, section, name);

Maybe plongest() instead of %llx?  Or does plongest not do the leading 0 fill
you want?

-- 
John Baldwin