On Sun, 14 Nov 2021, Hans Hagen wrote:
On 11/14/2021 4:33 AM, Aditya Mahajan via ntg-context wrote:
Hi,
There is a bug in lmt_arrow. If given a "non-cycle" path, lmt_arrow completes the cycle.
\starttext
\startMPcode newpath p; ahfactor := 5; p := origin .. (1cm, 1cm) .. (2cm, 0); draw lmt_arrow [ path = p, location = "percentage", percentage = 50, alternative = "curved", ] withpen pencircle scaled 1bp; \stopMPcode \stoptext
With 2021.11.11, I get the attached file. Note that lmt_arrow has drawn the "cycle" version of the path, even though p does not have a cycle.
I have looked at the code and cannot figure out why this is happening.
it happens in the 'direct' variant (metapost.registerdirect) where a boolean is passed as signal and that boolean then is also picked up by the (shared) path injector as (old school) signal that a path is closed while it should instead consutl the closed field in a path table (sounds more complex than it is)
so in the end it is a nasty side effect of the fast inject code path (shared function with different stack values; i'll probably have to ditch a "low level backward compatibility feature no one uses anyway" some day)
i uploaded a fix that you can test when the bins are compiled at
https://build.contextgarden.net/#/waterfall?tags=c.luametatex
new bins then in:
https://dl.contextgarden.net/build/luametatex/
(fyi: in the mp code you see that the path p is 'get' from the lua end where it got passed to by parameter)
Thanks! Works perfectly here.
(interesting that it didn't show up sooner, maybe most users have closed paths and/or most lmt_ features used closed paths)
I was reusing two year old code and took me a while to figure out why the output was suddenly wrong. Aditya