440724 - Folder tree couldn't open/close Thunderbird compiling with Sun Studio 12 x86

Status: RESOLVED FIXED

(Core :: XUL, defect)

Description

[Ginn Chen]

Compile Thunderbird with Sun Studio 12 x86.
The folder tree in the left couldn't be opened or closed.
Actually there's no twisy-open.png or twisty-clsd.png displayed.

This is a bug of Sun Studio 12, it doesn't allow "Assigning a negative value to an implicit unsigned bitfield."

So this code could not work:
 63     // N.B. these values are chosen to avoid problems with
 64     // sign-extension from the bit-packed Row structure.
 65     enum ContainerType {
 66         eContainerType_Unknown      =  0,
 67         eContainerType_Noncontainer =  1,
 68         eContainerType_Container    = -2
 69     };
 90     struct Row {
 91         nsTemplateMatch* mMatch;
 92         ContainerType    mContainerType  : 2;
 93         ContainerState   mContainerState : 2;
 94         ContainerFill    mContainerFill  : 2;
 95         
 96         Subtree*         mSubtree; // XXX eventually move to hashtable
 97     };

Compiled with current SS12, we will get mContainerType as "+2" not "-2".
The bug is logged to SS12.

According to
9.6p4 in the C++ standard:
    If the value of an enumerator is stored into a bit-field of the
    same enumeration type and the number of bits in the bit-field
    is large enough to hold all the values of that enumeration
    type (7.2), the original enumerator value and the value of the
    bit-field shall compare equal.

We can use 0,1,2 and pack it to 2bits safely, and get around SS12's bug.

Comment 1

[Ginn Chen]

Attachment: patch

Comment 2

[[e-mail]]

(In reply to comment #0)
> This is a bug of Sun Studio 12, it doesn't allow "Assigning a negative value to
> an implicit unsigned bitfield."
Why does it think the bitfield is unsigned? VC7.1 says it's signed.

Comment 3

[Ginn Chen]

(In reply to comment #2)
> (In reply to comment #0)
> > This is a bug of Sun Studio 12, it doesn't allow "Assigning a negative value to
> > an implicit unsigned bitfield."
> Why does it think the bitfield is unsigned? VC7.1 says it's signed.
> 

I think it's a bug of the compiler.
There might be some other compilers which have the same problem.

Comment 4

[[e-mail]]

OK, so VC7.1 (and VC6, since I had it handy) always compiles enum as signed; gcc however (since at least 2.96) compiles it as unsigned unless it explicitly contains negative values, while Sun Studio seems to think it's always unsigned!

Comment 5

[[e-mail]]

I think the spec is a $$$ document so someone else will have to chip in here, but I Googled around and found that ARM thinks that enums must be signed, while IBM seems to think that they will be signed unless you use large integers.

Comment 6

[Ginn Chen]

I guess the most portable solution here is to use {0,1,2} and packed into 4-bit.
It won't change the size of struct Row, I guess it's at least 16-bit aligned.

Comment 7

[Ginn Chen]

Attachment: patch v2

Comment 8

[Ginn Chen]

Pushed

http://hg.mozilla.org/index.cgi/mozilla-central/rev/65b3e7e453a6