Re: md5 lossage

[Dave Love]

Richard Stallman <address@hidden> writes:

> The file md5.c has not changed since the start of 2001, so if it
> did not have an endian problem in 21.3, I doubt it has one now.
> It seems to be designed to cope with endian issues.

I don't think it's in the basic code, since that's essentially the
same as what's in libc, though the (older?) version that md5sum uses
is somewhat different.  It must be something subtle.

> I have a vague memory that the Sun and MIPS are 64-bit machines at
> least recently.  Could this be a 64-bit problem, perhaps?

They are 64-bit CPUs (in sufficiently recent incarnations), but this
is in the default 32-bit userland.

> If you could step through the code in the case that fails on both the
> Sun and the x86, comparing the data at various points, I expect you
> could find the bug fairly straightforwardly.  Could you please
> do that?

I already had a go, and it wasn't straightforward for reasons I
forget.  I'll try again sometime when I can find time if there's
really no-one else who can debug on non-x86/lignux.