gdb 8.0 "lazy_string" exception "Length is larger than array size"

gdb 8.0 "lazy_string" exception "Length is larger than array size"

[Michael Stahl]

hi,

for the following string type:

  /** The implementation of a Unicode string.
  */
  typedef struct SAL_DLLPUBLIC_RTTI _rtl_uString
  {
      oslInterlockedCount refCount; /* opaque */
      sal_Int32           length;
      sal_Unicode         buffer[1];
  } rtl_uString;

the gdb python pretty-printer calls:

	return data.lazy_string(encoding, length)

full python pretty-printer module:

https://gerrit.libreoffice.org/gitweb?p=core.git;a=blob;f=solenv/gdb/libreoffice/util/string.py;h=32583718f83b2ad5707f75dd6327d9aa62764439;hb=5f210715fe090b4db4c80dcdee5f77dc404cf85c#l56

now this results in this exception:

  Traceback (most recent call last):
    File "/work/lo/master/solenv/gdb/libreoffice/util/string.py", line
29, in to_string
      return self.make_string(data, self.encoding, len)
    File "/work/lo/master/solenv/gdb/libreoffice/util/string.py", line
66, in make_string
      return data.lazy_string(encoding, length)
  gdb.error: Length is larger than array size.

this is with Fedora 26 "GNU gdb (GDB) Fedora 8.0.1-26.fc26" - in Fedora
25 this did not throw an exception.

apparently the problem is that the array is statically declared as
"buffer[1]", however its actual dynamic size is the same as "length".

is this a bug in gdb or is lazy_string not intended to support this
scenario?

regards,
 michael

Re: gdb 8.0 "lazy_string" exception "Length is larger than array size"

[Phil Muldoon]

On 25/09/17 18:20, Michael Stahl wrote:
> 
> hi,
> 
> for the following string type:
> 
>   /** The implementation of a Unicode string.
>   */
>   typedef struct SAL_DLLPUBLIC_RTTI _rtl_uString
>   {
>       oslInterlockedCount refCount; /* opaque */
>       sal_Int32           length;
>       sal_Unicode         buffer[1];
>   } rtl_uString;
> 
> the gdb python pretty-printer calls:
> 
> 	return data.lazy_string(encoding, length)
> 
> full python pretty-printer module:
> 
> https://gerrit.libreoffice.org/gitweb?p=core.git;a=blob;f=solenv/gdb/libreoffice/util/string.py;h=32583718f83b2ad5707f75dd6327d9aa62764439;hb=5f210715fe090b4db4c80dcdee5f77dc404cf85c#l56
> 
> now this results in this exception:
> 
>   Traceback (most recent call last):
>     File "/work/lo/master/solenv/gdb/libreoffice/util/string.py", line
> 29, in to_string
>       return self.make_string(data, self.encoding, len)
>     File "/work/lo/master/solenv/gdb/libreoffice/util/string.py", line
> 66, in make_string
>       return data.lazy_string(encoding, length)
>   gdb.error: Length is larger than array size.
> 
> this is with Fedora 26 "GNU gdb (GDB) Fedora 8.0.1-26.fc26" - in Fedora
> 25 this did not throw an exception.
> 
> apparently the problem is that the array is statically declared as
> "buffer[1]", however its actual dynamic size is the same as "length".
> 
> is this a bug in gdb or is lazy_string not intended to support this
> scenario?
> 
> regards,
>  michael
> 

Yeah, it's determining the string is an array and finding that the
declared length is larger then the array size. This is (as you noted)
the initial length of the array is [1] and that array is then
modified later. This code was added at 34b433203b5 by Doug Evans and
it was noted it was a bug. I've not sure, though, fixing this bug
may have had unintended consequences. I've CC'd Doug on the patch
and maybe he could comment further.

Cheers

Phil

Fwd: Re: gdb 8.0 "lazy_string" exception "Length is larger than array size"

[Phil Muldoon]

Oops noticed original email was addressed to gdb@gnu.org. I don't
think this is an alias for gdb@sourceware.org so forwarding my reply
for completeness.

Cheers

Phil

-------- Forwarded Message --------
Subject: Re: gdb 8.0 "lazy_string" exception "Length is larger than array size"
Date: Tue, 26 Sep 2017 08:56:03 +0100
From: Phil Muldoon <pmuldoon@redhat.com>
To: Michael Stahl <mstahl@redhat.com>, gdb@gnu.org, Doug Evans <dje@google.com>, Doug Evans <xdje42@gmail.com>

On 25/09/17 18:20, Michael Stahl wrote:
> 
> hi,
> 
> for the following string type:
> 
>   /** The implementation of a Unicode string.
>   */
>   typedef struct SAL_DLLPUBLIC_RTTI _rtl_uString
>   {
>       oslInterlockedCount refCount; /* opaque */
>       sal_Int32           length;
>       sal_Unicode         buffer[1];
>   } rtl_uString;
> 
> the gdb python pretty-printer calls:
> 
> 	return data.lazy_string(encoding, length)
> 
> full python pretty-printer module:
> 
> https://gerrit.libreoffice.org/gitweb?p=core.git;a=blob;f=solenv/gdb/libreoffice/util/string.py;h=32583718f83b2ad5707f75dd6327d9aa62764439;hb=5f210715fe090b4db4c80dcdee5f77dc404cf85c#l56
> 
> now this results in this exception:
> 
>   Traceback (most recent call last):
>     File "/work/lo/master/solenv/gdb/libreoffice/util/string.py", line
> 29, in to_string
>       return self.make_string(data, self.encoding, len)
>     File "/work/lo/master/solenv/gdb/libreoffice/util/string.py", line
> 66, in make_string
>       return data.lazy_string(encoding, length)
>   gdb.error: Length is larger than array size.
> 
> this is with Fedora 26 "GNU gdb (GDB) Fedora 8.0.1-26.fc26" - in Fedora
> 25 this did not throw an exception.
> 
> apparently the problem is that the array is statically declared as
> "buffer[1]", however its actual dynamic size is the same as "length".
> 
> is this a bug in gdb or is lazy_string not intended to support this
> scenario?
> 
> regards,
>  michael
> 

Yeah, it's determining the string is an array and finding that the
declared length is larger then the array size. This is (as you noted)
the initial length of the array is [1] and that array is then
modified later. This code was added at 34b433203b5 by Doug Evans and
it was noted it was a bug. I've not sure, though, fixing this bug
may have had unintended consequences. I've CC'd Doug on the patch
and maybe he could comment further.

Cheers

Phil

Re: gdb 8.0 "lazy_string" exception "Length is larger than array size"

[Pedro Alves]

On 09/26/2017 08:56 AM, Phil Muldoon wrote:
> On 25/09/17 18:20, Michael Stahl wrote:

> This code was added at 34b433203b5 by Doug Evans and
> it was noted it was a bug. I've not sure, though, fixing this bug
> may have had unintended consequences. I've CC'd Doug on the patch
> and maybe he could comment further.

We could perhaps decide to special case trailing arrays of
lengths 0 and 1 (i.e., let the caller request more elements than
declared), assuming they're being used as the trailing array
idiom, similarly to how gcc also has special concessions for those.
I don't know off hand whether its easy for the gdb code in question
to tell whether the array is the last field of a struct, though
I'd assume not.

If you want to ignore the array's declared length, I think you can
always decay 'buffer' to a pointer and work with that, and then GDB
won't have a length to validate.

Thanks,
Pedro Alves

Re: gdb 8.0 "lazy_string" exception "Length is larger than array size"

[Michael Stahl]

On 26.09.2017 12:30, Pedro Alves wrote:

> If you want to ignore the array's declared length, I think you can
> always decay 'buffer' to a pointer and work with that, and then GDB
> won't have a length to validate.

thanks, replacing
 ['buffer']
with
 ['buffer'].address
in the pretty printer avoids the exception!