Bug 21675

Summary:	Regression since 8.0 printing has changed format for decimal/octal/others
Product:	gdb	Reporter:	Jonah Graham <jonah>
Component:	gdb	Assignee:	Tom Tromey <tromey>
Status:	RESOLVED FIXED
Severity:	normal	CC:	pedro, tromey
Priority:	P2
Version:	HEAD
Target Milestone:	8.1
Host:		Target:
Build:		Last reconfirmed:	2017-07-11 00:00:00

Description Jonah Graham 2017-06-26 13:07:16 UTC

For example on current HEAD:

$ gdb/gdb -q -ex 'print/o -5' -ex quit
$1 = 037773777773

Instead previous versions do this:

$ gdb.8.0 -q -ex 'print/o -5' -ex quit
$1 = 037777777773

AFAICT This is a regression caused by d9109c8080ac0a2caeaa5f3d51e8a3b4891a68a8 (Simplify print_scalar_formatted)

There are many other examples of things that fail in a similar way. They were picked up by the DSF-GDB testsuite for Eclipse CDT. https://ci.eclipse.org/cdt/view/Debug/job/debug-tests-master-gdb-master

Comment 1 Jonah Graham 2017-06-26 13:30:21 UTC

Here is another example, demonstrating the problem is not just octal printing:

$ gdb/gdb -q -ex 'print/d 1.5' -ex quit
$1 = 1.5

$ gdb.8.0 -q -ex 'print/d 1.5' -ex quit
$1 = 1

and

$ gdb/gdb -q -ex 'print/u 1.5' -ex quit
$1 = 4609434218613702656

$ gdb.8.0 -q -ex 'print/u 1.5' -ex quit
$1 = 1

Comment 2 Tom Tromey 2017-07-11 02:34:05 UTC

(In reply to Jonah Graham from comment #1)
> Here is another example, demonstrating the problem is not just octal
> printing:

> $ gdb.8.0 -q -ex 'print/d 1.5' -ex quit
> $1 = 1

> $ gdb.8.0 -q -ex 'print/u 1.5' -ex quit
> $1 = 1

These still seem weird to me but I think we agreed to try to preserve
this in:

https://sourceware.org/ml/gdb-patches/2017-06/msg00314.html

Comment 3 Tom Tromey 2017-07-11 12:25:40 UTC

(In reply to Jonah Graham from comment #0)
> For example on current HEAD:

> $ gdb/gdb -q -ex 'print/o -5' -ex quit
> $1 = 037773777773

This bug is that one of the constants in print_octals_chars is wrong.
I guess it's been wrong for years and years and nobody noticed before...

Comment 4 Pedro Alves 2017-07-11 12:56:38 UTC

(In reply to Tom Tromey from comment #2)
> (In reply to Jonah Graham from comment #1)
> > Here is another example, demonstrating the problem is not just octal
> > printing:
> 
> > $ gdb.8.0 -q -ex 'print/d 1.5' -ex quit
> > $1 = 1
> 
> > $ gdb.8.0 -q -ex 'print/u 1.5' -ex quit
> > $1 = 1
> 
> These still seem weird to me 

OOC, how would you see them ideally be printed?

> but I think we agreed to try to preserve this in:

> https://sourceware.org/ml/gdb-patches/2017-06/msg00314.html

Comment 5 Tom Tromey 2017-07-11 15:35:35 UTC

(In reply to Pedro Alves from comment #4)

> > https://sourceware.org/ml/gdb-patches/2017-06/msg00314.html

For floating point types, I think /d and /u should work just like
not having a flag; /o should probably just be an error or ignored,
and /x should print in hex float.  I'm not sure what /t should do.

Comment 6 Pedro Alves 2017-07-11 19:58:52 UTC

Given structs with mixed types, it seems to me that erroring wouldn't be desirable.  E.g.:

  struct { int i; float f; } var;
  (gdb) p /o var

Comment 7 Tom Tromey 2017-07-12 02:19:32 UTC

(In reply to Pedro Alves from comment #6)
> Given structs with mixed types, it seems to me that erroring wouldn't be
> desirable.  E.g.:
> 
>   struct { int i; float f; } var;
>   (gdb) p /o var

Good point.  Maybe /o and /t could be ignored for floating point.
My main concern there would be possible confusion (like, "p/t 1.1"
could be weird) -- though on the other hand I think the current
behavior is already very confusing.

Comment 8 Jonah Graham 2017-08-16 09:47:16 UTC

Thanks for fixing this, the test failures in Eclipse CDT are now resolved.

PS due to https://sourceware.org/ml/overseers/2017-q3/msg00065.html some comments are missing in this bug:
https://sourceware.org/ml/gdb-prs/2017-q3/msg00203.html
https://sourceware.org/ml/gdb-prs/2017-q3/msg00204.html

Comment 9 Tom Tromey 2017-08-16 13:06:56 UTC

Thanks for posting those links.

The bug is fixed, I'm closing this.