CFF warnings

[behdad]

../../src/hb-ot-cff1-table.hh:374:60: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       if (unlikely (!ranges[i].sanitize (c) || (num_glyphs < ranges[i].nLeft + 1)))
                                                ~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~

Is there reason that nLeft etc are signed?

[blueshade7]

nLeft is HBUINT8 or HBUINT16 so must be unsigned. Any chance +1 makes it signed? Weird.

[behdad]

Right. Yes. That's C rules. Make it +1u.

[behdad]

nLeft is HBUINT8 or HBUINT16 so must be unsigned. Any chance +1 makes it signed? Weird.

What happens is that HBUINT8 and HBUINT16 have cast operators to uint8_t and uint16_t respectively. When those are added with 1 which is int, the common type that is chosen is int, since it can fit both values...

[blueshade7]

Ah thanks for reminding me. For this particular case I guess I would replace the condition with num_glyphs <= ranges[i].nLeft. More concise and potentially more efficient on some platform.

[behdad]

Ah thanks for reminding me. For this particular case I guess I would replace the condition with num_glyphs <= ranges[i].nLeft. More concise and potentially more efficient on some platform.

"<=" if done in loops causes the unsafe-optimization warning. But in conditionals it is indeed more concise and efficient.

Maybe I should change the cast operators for HBUINT8/16/... to int and unsigned int, to avoid this issue. Let me work on that.

[behdad]

Let me change the cast. I think you don't need to change.

[behdad]

Okay, that did it.

[behdad]

For this particular case I guess I would replace the condition with num_glyphs <= ranges[i].nLeft. More concise and potentially more efficient on some platform.

Please do this, as that's better indeed.

My other fix attempt broke things. Investigating.