Developer Meeting Agenda 2026-10-11 - 10:00 CEST

[grandixximo]

According to our bi-weekly schedule alternating between "early" and "late", this video meeting is Sunday 2026-Oct-11 at 10:00 CEST.

https://greenlight.bbb.uni-rostock.de/b/ste-c4d-brs-3k6
Access code: 869782

https://www.timeanddate.com/worldclock/meetingdetails.html?year=2026&month=10&day=11&hour=8&min=0&sec=0&p1=319&p2=236&p3=240&p4=136&p5=165&p6=256

Agenda

1. UI communication and the NML replacement: Zmq halui gui #4613, WIP: halui: pin-only GUI bridge surface (halui side of the #4613 split) #4616, discussion NML replacement #4571, issue Unify the way MDI commands are executed from HAL / GUI #4587. Zmq halui gui #4613 by @c-morley adds a ZMQ channel from halui to the GUIs, with Python embedded in halui. Review raised the embedded interpreter, a port clash with hal_bridge, and @rene-dev's objection that this puts ZMQ on top of NML instead of replacing it; his plan in NML replacement #4571 is to move halui into task first, then replace NML with flatbuffers over websockets. WIP: halui: pin-only GUI bridge surface (halui side of the #4613 split) #4616 (draft) sketches a split: halui only exposes plain HAL pins, and a userspace comp speaks the ZMQ. @BsAtHome on WIP: halui: pin-only GUI bridge surface (halui side of the #4613 split) #4616: every patch adds another layer (NML, HAL, ZMQ), HAL was meant as the RT interface and is turning into a polled message bus, and there is still no written list of requirements for the NML replacement. Since then Nml cleanup #4627 (step 1, NML out of the places where it is only a helper) is merged and move Halui into task #4644 (step 2, halui into task) is open as a draft. In NML replacement #4571 the talk moved to transport: websockets as the one transport so browser clients need no bridge (@rene-dev, @rmu75), or an interface independent of the transport (@BsAtHome); @alex-pres now agrees websockets fit the UI side. @BsAtHome objects that code is landing before a plan has consensus, and that step 1 left logging undecided (item 12). Since 10-07: @rmu75 posted a problem statement and a five-step plan in NML replacement #4571 (direction and requirements, then the communication and serialization choice, then a 1:1 replacement, then new remote APIs, then a multi-instance interpreter); @BsAtHome wants the architecture and the responsibilities defined before any technology is picked; @alex-pres listed the channels a web UI needs (he cannot attend); I proposed no further NML-related merges until a short written plan exists. Gui panel pins #4653 (@rene-dev, draft) is a third take on the Zmq halui gui #4613 panel pins: pins created through gstat, the same in every UI that uses gstat. Questions:
   • Requirements for the NML replacement: who writes the list, and where (NML replacement #4571, or a new issue)?
   • Order: halui into task first, then panel-to-GUI features on top of it?
   • Until then, do we accept a stopgap like Zmq halui gui #4613 or WIP: halui: pin-only GUI bridge surface (halui side of the #4613 split) #4616, and in which shape, or hold panel-to-GUI features?
   • Multiple UIs: @rene-dev wants a single source of truth in task for state such as selected axis, jog rate and increment. Does that cover what Zmq halui gui #4613 needs?
   • Unified MDI execution from HAL and the GUIs, Unify the way MDI commands are executed from HAL / GUI #4587.
   • Plan before code: do we agree the steps, and what each one leaves open, before more of it lands?
2. halui -> task @rene-dev move halui functionality to task
3. 32-bit userspace, libnml: use int64_t instead of long for CMS/NML sizes and ids #4614 (moved over from the hal: Make halcmd 64-bit clean after HAL API break #4611 thread). 32-bit was deprecated in 2.9.1 and 2.8 was the last release that supported it. An i386 build of master fails in one place only, libnml (260 errors from efe4ef2); libnml: use int64_t instead of long for CMS/NML sizes and ids #4614 fixes it in 18 lines, and it is back to draft until this is decided. In the thread @rmu75 and @rene-dev are fine with it going in as is, since libnml is going away; @rmu75 notes that without a 32-bit CI runner 32-bit breakage will keep creeping in; @BsAtHome does not want the (long long) print casts and asked for a move to fmt, which hal: print 64-bit values and pointers without long long casts #4617 does for the remaining HAL print sites (PRId64/PRIxPTR, no casts). Decision wanted: do we still want the tree to build on ILP32, and if so, do we keep new code 32-bit clean in review, or do we declare LP64 only and stop checking?
4. HAL 64-bit, phase three. hal: Break the HAL API - Move to 64-bit exclusively #4565 is in (2026-09-27), the docs followed in docs: Update to reflect the HAL API changes #4596 to docs: Update man9 pages to reflect the HAL API changes. #4599, and v2.10.0-pre2 is tagged. What is left:
   • Remaining 32-bit getters and setters: hal: Make hal_gm and hal_ppmc drivers 64-bit clean after HAL API break #4609 (hal_gm, hal_ppmc), hal: Make .comp components 64-bit clean after HAL API break #4610 (.comp components) and hal: Make halcmd 64-bit clean after HAL API break #4611 (halcmd) are in. 57 files still use the 32-bit accessors, some on purpose (32-bit hardware counts, as in hal_gm). Do we want a list of what still needs migrating, and owners?
   • hal: C++ API (hal.hh) and pybind11 bindings on the new HAL API #4251, C++ API (hal.hh) and pybind11 bindings on the new API.
   • Draft - HAL types update - Remove 32-bit and move to 64-bit #4247, the draft the break started from. Close it now that hal: Break the HAL API - Move to 64-bit exclusively #4565 is merged?
   • Converters: hal: Add halfileupdate to convert HAL configuration files #4499 (halfileupdate) was draft until the renaming settled, and hal: Break the HAL API - Move to 64-bit exclusively #4565 settled it. hal: Add halcompupdate to migrate .comp files to the new HAL API #4256 (halcompupdate) conflicts after the break. Users are on the new API now without either converter; can these go in next?
   • hm2_modbus .mbccb files need recompiling from .mbccs after the break. Last meeting left open where we tell users: release notes, NEWS, or docs?
5. Signed vs unsigned integers in HAL, discussion The unsigned integer question #4526. Conclusion last time: "In principle alright. Needs documentation and some extra discussion." What is left is assigning the pieces:
   • Static conv_* components with replace_nan / replace_inf_positive / replace_inf_negative pins, and no clamp on the real to int direction (needs .hal file updates).
   • Cross-netting bool/sint/uint as a warning instead of an error, which also fixes hal_get_bool() reading false on big-endian.
   • Documentation of the decision. Who writes which piece?
6. Development model, discussion Branch Protection #4558. Conclusion last time: CI enforcement off, branch protection off for now, needs discussion. Part of the friction (see @snowgoer540's comment in Developer Meeting Agenda 2026-09-27 - 20:00 CEST #4566) is that the model was never written down, so rules got applied that nobody had agreed to. Rather than argue past cases, let's define the current model concretely:
   • Areas and maintainers: which parts of the tree have a maintainer (GUIs have worked that way for years), and who are they?
   • What may be committed directly and what goes through a PR: by area, by size, or by whether it touches shared code (HAL, motion, interpreter, build, CI)?
   • Changes that cross areas: whose review is needed when a change in one area breaks another?
   • CI: does a PR need green CI to merge, and what is the escape hatch for infrastructure flakes?
   • Branch protection: once the above is agreed, configure GitHub to match it and not the other way round.
   • Changing the model: announced on the list or in a discussion and agreed before it is enforced, never retroactively.
   • Where it is written down: a page in the Developer Manual?
7. Include layout. hal: Break the HAL API - Move to 64-bit exclusively #4565 is in, so per the last two meetings include: move the exported headers into a directory of their own #4461 goes next and then the angle-bracket check include: use angle brackets for exported headers, and check it #4375. @BsAtHome's git switch question on include: move the exported headers into a directory of their own #4461 is answered (silent both ways), and it is rebased on current master. Merge?
8. Versioning and release procedure, CI build artifacts on master have wrong version #4555 and Rework versioning scripts and workflow #4584. Conclusion last time: the direction is right, details need discussion. Since then the v2.10.0-pre2 tag exists, while VERSION_BASE says 2.10.0~pre2, so the tilde/dash question now applies to a real tag. Open points from last time: strictness on an untagged VERSION_BASE change, non-release tags, source packages built from a zip, and the backport to 2.9. Rework versioning scripts and workflow #4584 is still draft and conflicts.
9. ini identifiers. Decided last time: keep the strict [a-zA-Z_][a-zA-Z0-9_]*. inifile: accept dash in section and variable identifiers #4573 is closed; xhc-hb04: rename dashed ini identifiers, migrate old configs in update_ini #4601 renames the dashed xhc-hb04 identifiers and migrates old configs in update_ini. Once it is in, the Tcl flattening in Remove Tcl from sample configs #4524 can start with the pendant configs.
10. [DISPLAY]CYCLE_TIME, Unify [DISPLAY]CYCLE_TIME unit handling across GUIs #4424. Decided on 2026-09-13: every GUI/VCP reads the value the gmoccapy way first. No PR yet. Who takes it?
11. Status checks on owners named last meeting:
    • hm2_eth read timeouts, hm2_eth read timeout #4525: Rene/Robert/Andy, configs and logs from real hardware.
    • Homing, Andy: homing: stop the correct joints when aborting a homing sequence #4574 and homing: bound the homing moves, return from a failed latch move #4578 are merged. homing: run every leg of a synchronized home sequence together #4577 (every leg of a synchronized sequence together) is open; Andy's question on misaligned switches compensated by HOME_OFFSET is answered in the thread. HOME_OFFSET is doing double duty #4527 (HOME_OFFSET double duty) has a design proposal now, see item 15. Add Support for Dogbone Homing #4009 dogbone homing conflicts with master.
    • hostmot2, fix safety and correctness bugs in hostmot2 drivers #4488: the split is posted (sserial + abs_encoder, pktuart, eth, the rest); waiting on @sleepybishop.
12. Logging library #4638 (optional if there is time) currently a bunch of log messages get converted from printf style format specifiers to fmtlib type safe formats, in the course of these changes, it would be easy to clean up logging and introduce a proper logging library. opened github discussion to gather requirements and ideas.
13. plan release 2.10: review what remains to be done
    • Mailing list thread "Will we release before more major changes?" (2026-10-09): @c-morley asks for a release before the NML work, since big changes take 6 to 12 months to settle. @andypugh and @rmu75 propose branching 2.10 off master once the HAL changes are done (a new stable branch, as in the ReleaseCheckList wiki page), before NML and the new TP. @BsAtHome asks not to split yet: the tree is not 64-bit clean (hostmot2, 64-bit encoder counters), and he prefers a feature freeze on master over a branch, so testing is not split over two trees. NML replacement is 3.0 material.
    • Decide: branch now, freeze master first, or branch once a named list is done, and which list?
14. mc_axis proposal, discussion Proposal: mc_axis — a PLCopen MC Part 1 axis component for HAL (extra joints, axes without G-code) #4647 by @yurc: a PLCopen Motion Control Part 1 axis as a HAL component (power, absolute/relative/velocity moves, halt, set-position, the PLCopen state machine, following error and drive fault handling, optional CiA-402 profile position), for extra joints and HAL-only axes. Field-tested on an EtherCAT servo; tests and a sim config are on the branch. @andypugh points at the overlap with limit3 and simple_tp. Do we take it into the tree, and under which name? Related: should the cia402 component move from linuxcnc-ethercat into the tree? Update: @yurc added a track mode and a virtual axis and a side-by-side test against simple_tp, so mc_axis could replace the simple_tp component (used only by the pncconf axis test); limit3 stays, it does a different job.
15. HOME_OFFSET double duty, issue HOME_OFFSET is doing double duty #4527, follow-up from the 2026-09-13 meeting. In HOME_ABSOLUTE_ENCODER mode HOME_OFFSET is the encoder zero, everywhere else the switch position, so a machine with an absolute encoder and home switches cannot express both. Proposal in the issue, in three phases:
    • Phase 1: new optional [JOINT_n]HOME_ENCODER_OFFSET, falling back to HOME_OFFSET, so existing configs behave the same.
    • Phase 2: a task-written state file holding the learned encoder reference (same precedent as the VAR file and POSITION_FILE), an opt-in HOME_ENCODER_TOLERANCE plausibility check at boot, re-reference through the home switch when the check fails, and a pin showing where the position came from.
    • Phase 3: G27, the reference position return check.
      @rene-dev asks for a solid test suite covering all encoder types first. Is the direction right, and are the parameter and file names acceptable?
16. HAL names, two issues from @dirkx:
    • Feature request: (Not)Accepting decoration as a hal signal name when there is no space #4645: net lube-pump=> ethercat.x creates a signal named lube-pump=>, because the decoration only counts with a space before it. Interactive halcmd does not accept such a name back, so the running config cannot be queried. Proposal: a name may not end in a decoration (=>, <=, <=>), so a missing space becomes an error; input=>output in the middle stays legal (@rmu75 notes that could still break a config). We chose strict identifiers for the ini last meeting (item 9); do we want the same for HAL names, or only this rule?
    • hal - alphanumeric sort #4650: natural sort (D9 before D10) in halcmd show, halshow and halscope, which matters with EtherCAT's hundreds of unpadded names. @BsAtHome: the HAL library's internal order is not an interface; the tools should sort. Do we want sorting options in halcmd and the GUIs?
17. Meeting day and time: @rmu75 asks on the list for a more convenient weekday or time than Sunday morning.
18. PRs
    See PRs for merge/comments.
    • hm2_eth: Xenomai4 OOB improvements #4503 hm2_eth Xenomai4 OOB: rebased and re-tested by @hdiethelm on Sep 27.
    • change external offset to float #4521 change external offset to float
    • change time pins to float #4518 change time pins to float
    • sim_encoder: fix negative velocity at the thread rate cap, raise sim spindle speed limits #4583 sim_encoder negative velocity and sim spindle speed limits.
    • pumakins: keep joints 4 and 6 within a half turn of the previous value #4586 pumakins: keep joints 4 and 6 within a half turn of the previous value.
    • GUIs and preview: an axis converts by its [AXIS_<letter>] TYPE #4594 GUIs and preview: an axis converts by its [AXIS_<letter>] TYPE, follow-up to Honor [AXIS_<letter>] TYPE in the interpreter, canon and motion #4581.
    • switchkins: one dispatch for kinematics modules, halcompile components and out-of-tree modules #4595 switchkins: one dispatch for kinematics modules, halcompile components and out-of-tree modules.
    • interp: the tilted work plane, G68.2, G68.4 and G69 #4600 tilted work plane, G68.2, G68.4 and G69.
    • qtvcp: ignore traceback-less exceptions in excepthook #4612 qtvcp: ignore traceback-less exceptions in the excepthook.
    • libnml: use int64_t instead of long for CMS/NML sizes and ids #4614 libnml int64_t: see item 3.
    • hal: print 64-bit values and pointers without long long casts #4617 hal: print 64-bit values and pointers without long long casts, see item 3.
    • preview: one preamble for every GUI, restating modes only #4618 preview: one preamble for every GUI, restating modes only.
    • motion: check circular moves against axis soft limits #4619 motion: check circular moves against axis soft limits.
    • glnav: avoid divide-by-zero in set_eyepoint_from_extents #4620 glnav: avoid divide-by-zero in set_eyepoint_from_extents.
    • gomath: fix negative pitch singularities in RPY and ZYX conversion #4623 gomath: fix negative pitch singularities in RPY and ZYX conversion.
    • motion: make M62-M68 on a non-existent output an error #4624 motion: make M62-M68 on a non-existent output an error.
    • move Halui into task #4644 move halui into task (draft), see items 1 and 2.
    • homing: drop the dead HOME_DELAY timer in HOME_FINAL_MOVE_START #4629 homing: drop the dead HOME_DELAY timer in HOME_FINAL_MOVE_START.
    • hal: don't free a funct entry the running thread may still use (delf/unloadrt crash) #4630 hal: don't free a funct entry the running thread may still use (delf/unloadrt crash).
    • motion: per-axis homed check for teleop jog soft limits #4631 motion: per-axis homed check for teleop jog soft limits.
    • gcodemodule: report the actual exception from failed canon callbacks #4636 gcodemodule: report the actual exception from failed canon callbacks.
    • configure: fix libtirpc pkg-config detection #4639 configure: fix libtirpc pkg-config detection.
    • pyvcp: add an optional reset button to the spinbox widget #4640 pyvcp: add an optional reset button to the spinbox widget.
    • tp: make helix curvature independent of arc subdivision #4643 tp: circle curvature fix.
    • runtests: Add parallel mode #4648 runtests: parallel mode (draft).
    • Gui panel pins #4653 GUI panel pins through gstat (draft), see item 1.
    • qtplasmac: fix button highlight didn't update on color changes #4657 qtplasmac: fix button highlight not updating on color changes.
    • emcmodule: retry the tool mmap attach in stat.poll() #4658 emcmodule: retry the tool mmap attach in stat.poll().

[rmu75]

What is the process to get something on the agenda? I can edit it, but it seems stuff gets removed afterwards.

The HALUI / Task situation should be discussed IMO, also the external offsets float thing, in a sense that is more important than odd corner cases like sim_encoder or the 32bit fixes in NML.

[grandixximo]

Sorry, I mistakenly pasted back my local copy, go ahead with edits, I'll be more careful.

[grandixximo]

I was able to restore your edits from the edit history: the halui -> task item is now item 2, and #4521 is back in the PR list. Sorry again, I should have known better, since this post is meant to be edited by everyone.

[rmu75]

no problem. thanks. I will add further things as separate comments, I don't want to mess up your summaries.

[grandixximo]

I did the messing up, you go ahead edit the body, it was my bad

[andypugh]

Conclusions: (We didn't get far through the agenda)

1. NML: Build a prototype, see how it goes. (pretty much done by Rene). it's not too late to change the protocols, but at least it shows what needs to change to do this. Making the change is probably the best way to understand the change.... Evaluate ZMQ over WS

2. Move HALUI into task, do it.

3. Keep 32bit architecture compatible for now, at least for Debian 14 forky (beagle bone is very popular and 32bit), build needs fixing.

4. needs to go in before branching off 2.10, including C++ pybind11 python bindings. hal: C++ API (hal.hh) and pybind11 bindings on the new HAL API #4251

5. can prob. go in

6. postponed

---

9. requested review again from Bertho

10. @grandixximo is taking it

11. hm2 testing in progress, homing needs better test harness, absolute encoders on gantry with homing switches.

---

16. input from Bertho needed

---

18. PRs

• timing change to float: change time pins to float #4518 merged.
• external offsets: change external offset to float #4521 changes ext offsets to float, plasmac pins also to float. sim configs need updateing, maybe also the GUI.
• pumakins: keep joints 4 and 6 within a half turn of the previous value #4586 merged
• GUIs and preview: an axis converts by its [AXIS_<letter>] TYPE #4594 merged
• switchkins: one dispatch for kinematics modules, halcompile components and out-of-tree modules #4595 merged
• interp: the tilted work plane, G68.2, G68.4 and G69 #4600 merged
• qtvcp: ignore traceback-less exceptions in excepthook #4612 to be merged
• libnml: use int64_t instead of long for CMS/NML sizes and ids #4614 to be merged
• motion: check circular moves against axis soft limits #4619 rebase on master
• glnav: avoid divide-by-zero in set_eyepoint_from_extents #4620 merged
• gomath: fix negative pitch singularities in RPY and ZYX conversion #4623 rebase on master
• motion: make M62-M68 on a non-existent output an error #4624 merged
• move Halui into task #4644 see 2) needs some time / comments
• homing: drop the dead HOME_DELAY timer in HOME_FINAL_MOVE_START #4629 merged
• hal: don't free a funct entry the running thread may still use (delf/unloadrt crash) #4630 leave to bertho to merge
• motion: per-axis homed check for teleop jog soft limits #4631 merged
• gcodemodule: report the actual exception from failed canon callbacks #4636 merged
• configure: fix libtirpc pkg-config detection #4639 merged, also needs to go into master
• pyvcp: add an optional reset button to the spinbox widget #4640 merged
• tp: make helix curvature independent of arc subdivision #4643 @grandixximo wants to test a bit more
• Gui panel pins #4653 get input from @c-morley
• qtplasmac: fix button highlight didn't update on color changes #4657 merged
• emcmodule: retry the tool mmap attach in stat.poll() #4658 merged

mark issues / features that should still go into 2.10 with label 2.10-candidate and bugs with 2.10-must-fix

@rene-dev wants to C++ standard change to C++23, did all the research, bookworm and trixie supported. PR coming.