Re: RFC: make verifier more lazy

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Robert" == Robert Schuster <theBohemian@gmx.net> writes:

Robert> this is my patch for PR #17021. It makes the verification step
Robert> more lazy and allows loading of classes where classes files
Robert> are referenced from which the bytecode cannot be found.

Robert> I wrote a couple of mauve (module verify) tests to make sure
Robert> that I introduced no regressions and prove that my fix works.

Excellent.

Robert> @Bryce: I sent my paperwork back to the FSF office already (~10 days).

We have to wait for them to receive it before we can check anything
in.

Robert> +// For strcpy
Robert> +#include <string.h>

This doesn't need a comment.

Robert> +              if ( self->data.klass == &java::lang::Object::class$ )

In our coding style there are no spaces after '(' or before ')'.

Robert> +          if ( klass->data.name->limit()[-1] == ';' ||
Robert> +               _Jv_isDerivedFromPrimitive(klass->data.name) )

Instead of _Jv_isDerivedFromPrimitive, can't you just look for a
leading '[' here?

I suppose we need to test this to make sure it works with class names
like 'I'.

Robert> +jboolean
Robert> +_Jv_equalUtf8Classnames (const Utf8Const *a, const Utf8Const *b)
Robert> +{
Robert> +  if ( _Jv_equalUtf8Consts(a, b) )
Robert> +    return true;

This is inefficient.  Couldn't we handle this check in the same loop
as the other others?

Robert> +  // Skips any leading L characters (even if they belong to the package
Robert> +  // name).
Robert> +  while ( ac == 'L' )
Robert> +    ac = UTF8_GET(aptr, alimit);

I think there can only be one leading 'L', and you only want to strip
it if the last character is a ';'.  You can "strip" the ';'s by
decrementing the corresponding limit variable.

Robert> +  friend jboolean _Jv_equalUtf8Classnames (const _Jv_Utf8Const*, const _Jv_Utf8Const*);

This goes over 80 columns and should be split at the ','.

Tom