Re: are global exception flags a resource bottleneck?

[Guillaume Melquiond <guillaume.melquiond@xxxxxxxxxxx>]

Le mardi 28 août 2007 à 17:17 -0700, David Hough 754R work a écrit :
> I have been thinking about ways of reducing the impact of exception flags
> on programs that don't use them - most programs - while not losing any
> of the capabilities they provide.

Here are a few comments on your suggested changes:

In section 9, you should avoid writing "most cases" and "usually". The
purple prose (or some references to other sections) should be part of
the text, so that the reader is not confused by these terms. Also, the
"should be raised" should be reverted back to "shall be raised", and an
additional sentence should explicitly states the special case of
"raiseNoFlag".

In section E.3, I find the term "synchronous" misleading. If I
understand correctly, the main block is executed till its completion, so
the exception handling is not synchronous at all, it is _delayed_
instead. Also, the text does not make it clear that the alternate block
is executed _only_ if an exception would have been raised while in the
main block.

As a side note, I feel the descriptions of the "alternate exception
code" modes are not clear. They should state that the execution is
supposed to resume at the end of the main block once the attached block
has been executed (be it asynchronous or delayed handling). Currently,
the only hint that it is the mandated behavior is because the
description of the second "transfer" mode states "no return possible".

Previous comments were purely editorial in nature. The following one is
not. I feel the "substituteExor" mode is overly restrictive. It could be
renamed to something like "substituteCopysign", and it would copy the
sign of any operation results, not just multiplication and division. But
perhaps there is a rationale for having a mode that only applies to
multiplication and division?

Best regards,

Guillaume