factor out aborts #579

[ekipan]

Thanks for build fix.

[ekipan]

Apparently the issue reference doesn't totally work if it's in the title so just for completeness sake:

See #579.

[jkotlinski]

Nice, thanks for contributing! I will look at it in detail later, and see how it fits in with other stuff. lör 14 dec. 2024 kl. 09:44 skrev ekipan ***@***.***>:

…

Apparently the issue reference doesn't totally work if it's in the title so just for completeness sake: See #579 <#579>. — Reply to this email directly, view it on GitHub <#581 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAY34OYZ4NUYSOWBKUUBXJL2FPVWHAVCNFSM6AAAAABTTKT7HSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNBTGAYDMNBYGU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[jkotlinski]

OK, this seems like an improvement. I am pondering various things now.

• Is the new ABORTS useful somewhere in DurexForth itself? I am kind of wary about adding new words. But it would feel better if there was some immediate use for it.
• Would it make sense to code golf it harder? I.e. add (abort") which does all the work in one word. Check condition, and either print error message and abort, or skip error message.

[ekipan]

That means every runtime-untaken abort" turns from a single 0branch into rstack, counting, stack juggling etc. A bit more work. Plus I feel like you'd also need to pull a factor from lits to make it worth it, adding 12 cycles of jsr rts to every runtime string literal. Something like this?

: (lits) 1+ count 2dup + 1- ;
: lits r> (lits) >r ;

: (s") '"' parse dup c, 
here swap dup allot move ;
: s" postpone lits (s") ; immediate

\ ...

: (abort") r> (lits) >r rot 
if rvs type cr abort then 2drop ;
: abort" postpone (abort") (s") ;
immediate

[jkotlinski]

Uhh, that seems like a bad idea of mine. It is better to use if .. then for the conditional, otherwise it gets too slow.

[jkotlinski]

How about merging the PR as is? But I would like one change, rename aborts to (abort"). Then it is clear that it is a helper word to abort", and not a part of the official API.
That is, it does not need to be documented, and it can be later removed or changed without being a breaking change.

[ekipan]

This of course does add more a bit more work in the case abort" is given a nonzero at runtime, but considering its purpose I'm sure that's okay. 12 cycles spent in the rare case a user program fails for a 9 bytes savings every time they compile the word.

[ekipan]

Or I guess only 6 cycles, since abort doesn't rts :P

[jkotlinski]

Looks good to me! Maybe at some point this will throw an exception instead. sön 15 dec. 2024 kl. 16:04 skrev ekipan ***@***.***>:

…

Or I guess only 6 cycles, since abort doesn't rts :P — Reply to this email directly, view it on GitHub <#581 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAY34O2UVFTOOLXIO2F3IWT2FWK6PAVCNFSM6AAAAABTTKT7HSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNBTHEYDMMZWG4> . You are receiving this because you commented.Message ID: ***@***.***>