10:31 <dnkl> brocellous: hmm, that should be relatively easy to fix

10:32 <dnkl> the VT parser discards invalid UTF-8

10:32 <dnkl> should be simple to instead insert a replacement character

10:43 <ovf> i like how every implementation discussed above does its own thing with utf8

13:21 <dnkl> brocellous: the reason for the "lost" character is that E9 is valid (it indicates a three-byte UTF-8 sequence). So the invalid sequence is detected at 'm'

13:21 <dnkl> ovf: since the sequence is invalid, all above variants are valid, in a way...

13:24 <dnkl> I suppose we could special case ASCII characters; i.e. if an invalid UTF8 sequence is terminated by an ASCII character, print the ASCII character instead of dropping it as part of the invalid UTF8

13:24 <dnkl> but technically, what we're currently doing, _is_ correct

13:25 <dnkl> in a sense

13:27 <rockorager> the second byte shouldn’t be valid

13:28 <rockorager> it was ascii right?

13:28 <rockorager> but a 3 byte sequence would have bit 8 on

13:28 <dnkl> correct, e9 indicates a three byte sequence. 'm' is the second byte and is not a valid middle byte in a three byte utf8 sequence

13:29 <rockorager> oh I got it

13:29 <dnkl> foot's vt parser detects this and goes back to the 'ground' (default) state, while discarding the two utf8 bytes read so far

13:30 <rockorager> ok. I think for my tui libraries I throw away only one byte

13:30 <rockorager> this is what the go std lib does for invalid utf8 too

13:32 <dnkl> i suppose my suggestion above could be extended/modified to instead have the state machine go back to the default state, and _then_ feed it the "current" character ('m' in this case)

13:32 <dnkl> this would also let us handle longer utf8 sequences following an invalid one

13:36 <ovf> dnkl: yep, i think it's a failing of utf8 for not prescribing error recovery

13:38 <ovf> i agree that pok\xe9mon should be parsed as pok�mon (i.e. invalid octet terminates multi-byte sequence), because nobody uses serial links anymore, and cosmic rays aren't a huge problem yet (especially for a terminal emulator), so the probability that the illegal character is a result of a bit-flip is pretty low.

13:41 <dnkl> fwiw, invalid UTF-8, at all, should be very unusual, unless an application has mixed up its encodings...

13:42 <dnkl> https://codeberg.org/dnkl/foot/pulls/1999

16:27 <brocellous> Yeah I don't think there's really a "correct" behavior here. I guess my expectation comes from icu, since `uconv -t utf-8 --callback substitute <<< $'pok\xe9mon'` results in "pok�mon".

17:25 <aelius> I did end up dropping lanczsos3 scaling down to box (somewhat arbitrarily chosen)

17:25 <aelius> Because I cannot visually see the difference in the sample emoji.txt

17:27 <aelius> and dsepite being an unrealistic scenario, seems like I might as well save the cpu cycles :)

17:27 <aelius> (for the presence of any emoji)

17:29 <dnkl> aelius: that's what the tweak option is for :)

17:30 <dnkl> brocellous: I updated the PR (link above) to also insert a REPLACEMENT CHARACTER. With that, we should behave just like kitty

18:36 <aelius> With that configured, I think foot is pretty much perfect!

18:38 <aelius> This is basically what I had hoped the experience of suckless' st would be like- before I learned that their vision of "simple" involved requiring tmux/other :v

18:39 <j`ey> what does foot have that doesnt require tmux?

18:39 <aelius> a scrollback!

18:39 <aelius> lol

18:40 <j`ey> ah :D

18:44 <aelius> I can understand it, sort of- if a real terminal didn't have a scrollback (before my time)- sure. I guess. On the other hand, didn't og paper terminals technically have a "scrollback" because... it was printed on paper? :P

18:44 <aelius> Just seemed very non-suckless of suckless to say "here's our very simple low SLOC terminal- now put termux in it"

18:45 <aelius> er, tmux*

20:35 <llyyr> show-urls-persistent hits an assert on the latest release

20:35 <llyyr> > err: url-mode.c:113: BUG in spawn_url_launcher(): assertion failed: 'term->url_launch != NULL'

20:35 <llyyr> first url opens fine, but if you try to open a second url we hit that

20:36 <llyyr> ASAN also complains about this once on startup: ../grid.c:820:49: runtime error: variable length array bound evaluates to non-positive value 0

20:45 <llyyr> first bad commit for the crash is 051cd6ecfc8b98e1d80b399ebea40207fe040750

21:39 <ovf> aelius: can confirm, my hardcopy terminal has scrollback.

22:13 <aelius> :v

22:14 <aelius> (I did figure some models probably do)