490011 - abcasm exception handling is too hard to use and it needs something like try/catch

Status: VERIFIED FIXED

(Tamarin Graveyard :: Tools, defect)

Description

[Chris Peyer]

exception handling currently requires all exceptions to be defined in the .method_body_info element using from, to and target labels to specify where to deal with the exceptions.  This is not very readable as it is all defined outside of the code.  An as-like try/catch syntax would be much more readable and usable.

Comment 1

[Tom Harwood]

Possible syntax:

.try {
    instruction
    instruction
} .catch {
    getlocal0
    pushscope
    instruction
}

Per abcasm design goals, it's up to the assembly-language programmer to follow the rules (or not) regarding stack and scope depth, etc.

Comment 2

[Tom Harwood]

Attachment: try/catch support and fixes for two latent bugs

try/catch support, and fixes for two latent bugs.
try/catch is as in previous comment, with an enhancement: the catch type can be specified.
Latent bugs are:
- assembler blocks with no instructions weren't correctly handled
- stack depth processing didn't account for catch block's characteristics.

Comment 3

[Tom Harwood]

Attachment: Testcase and sample

Comment 4

[Tom Harwood]

Attachment: Patch without the binary diff

Comment 5

[Edwin Smith]

can you post an example of the abs for how this is used?

Comment 6

[Edwin Smith]

nevermind

Comment 7

[Edwin Smith]

Comment on attachment 374782 [details] [diff] [review]
Patch without the binary diff

looks fine as as long as it's still possible to do it the old/hard way (for testing situations you can't express this way, such as non-nested but overlapping try regions)

Comment 8

[Tom Harwood]

Yah, the old school way will continue to work and it has coverage; this is syntactic sugar for the "hard" way.  Semantic sugar we don't do by design, i.e., the assembler leaves balancing the stack in the hands of the assembly language programmer who will probably suffer a bit on that account.

Comment 9

[Tom Harwood]

Attachment: Revised patch without synthetic jump

Per F2F discussion with Ed, changed patch slightly: the to-be-pushed version doesn't insert a synthetic jump.  This makes the .try/.catch regions easier to modify into try/catch/finally, and generally explore more misadventures in control flow, which is a primary goal of the assembler :)

Comment 10

[Tom Harwood]

Attachment: Testcase/sample with explict jump/return out of try regions

Comment 11

[Tom Harwood]

Pushed patch and testcase: 1791:e1b14fa22f28