[kaffe] make check fails...

[Jeroen Frijters]

Andrew Haley wrote:
> Jeroen Frijters writes:
>
>  > However, the truth of the matter is that IMO our serialization code
>  > is broken beyond repair, so I don't think it's worth it to try to
>  > fix it the right way. We should just merge in the Sun serialization
>  > code once that is available ;-)
>
> This is very depressing.  I tweaked gcj's serialization code so that
> it worked correctly with a variety of code, including jonas.  This was
> a difficult and tedious job, to say the least.  The result was pretty
> robust, however.
>
> Now, gcj's own serialization code has been replaced with the Classpath
> code on the grounds that it was better tested, and had the bugs ironed
> out.  I haven't yet tried large-scale applications, but I have a bad
> feeling about this.
>
> If needs be we could just get libgcj's own serialization code back.

I don't know if that would be better. Classpath's code isn't exactly useless, it's just that the last 5% will be incredibly hard to get right (given the poor specification of serialization details). I suspect that gcj's version also had a lot of bugs in corner cases. That seems virtually inevitable in something this complicated that is so poorly specified (and wasn't even designed, but just evolved into what it is today).

As you can probably tell from the above, I'm no fan of serialization, so maybe that also colours my perception.

Regards,
Jeroen