Re: [Audacity-devel] Crash on launch

[James C. <cr...@in...>]

On 05/06/2018 23:37, Henric Jungheim wrote:
> TL;DR: Parameter type mismatches in variadic functions can
> cause trouble.
>
> As long as the size of the arguments match, and the numbers
> are small, then yes.  (Some pedants might point out that
> this assumes more than the language standard guarantees.  I
> would ask them to point to an example of an actual
> production CPU, that would otherwise be capable of running
> Audacity, where that is not a valid assumption.)
>
> More serious problem can arise depending on the argument
> passing conventions and the size of those arguments.  A
> little-endian number read as 32 bits when it really is 64
> bits should get the right output for that parameter, but may
> cause future parameters to be misaligned (e.g., treating the
> second 32-bits of the first number as part of the second
> parameter).  Getting things sufficiently wrong can cause
> segfaults and other such fun.  How bad things get can also
> depend on if an argument is on the stack or in a register
> (e.g., the first four parameters could be in registers, the
> rest on the stack).
>
> Do any of the supported platforms have a 32-bit size_t but a
> 64-bit long?  Do any have a 64-bit size_t, but 4-byte stack
> parameter alignment?  Are there any with a 64-bit size_t and
> a 32-bit long that are big endian?
>
> Windows on x86 should be okay because both size_t and long
> are 32-bits.
>
> Since Windows on x64 and ARM64 is little-endian, it should
> be okay as long as the actual value to be displayed is <
> 2^31.  The upper half of the register or stack location
> would be ignored.  Subsequent parameters would be fine,
> assuming that the stack parameters are aligned to at least 8
> bytes (and, IIRC, they are).
>
> The cross product of MIPS, ARM, and [insert your favorite
> obscure architecture here] with big-endian vs little-endian
> might find certain combination where this is not be okay.
>
> BTW, will Audacity 2.3.0 officially support any non-Intel
> architectures?  This would only be relevant for Linux,
> right?  People do compile for Raspberry PI, but is that
> officially supported or just "nice that it works"?

2.3.0 will only officially support Intel architecture.  That it compiles 
on a Raspberry PI is a 'nice that it works'.


> I did get Audacity to compile for Windows on ARM64 a while
> back, but I'll wait for the second generation devices later
> this year to see if it works (more likely, to see how it
> fails, and then try to get it working); the $700 for a
> first-gen Snapdragon 835 system exceeds my curiosity
> threshold.
>