fix(pcb): write the Default netclass complete, and repair one that is not - #333
Merged
mixelpixx merged 2 commits intoAug 29, 2026
Merged
Conversation
… not create_netclass wrote every class with four PCB fields and nothing else. For a class named "Default" that is not a partial description, it is a replacement: net_settings.cpp:216 swaps KiCad's seeded default out for whatever the file holds, and addMissingDefaults fills other classes *from* the Default while nothing fills the Default itself. Every key omitted there is left resolved from nothing. wire_width is the one that bites. Without it eeschema silently refuses to place a junction anywhere in the project, strips existing dots on the next save, and drops pins joined by a T to unconnected-*. There is no ERC violation and no error message. Plotted, the wire group comes out as "fill:none; stroke:none;" and the junction dot collapses from 0.4572 mm to 0.0001 mm. Which class is the default is decided by name: SetName sets m_isDefault on an exact match (netclass.h:96) and the loader branches on that alone. KiCad writing the Default first is an output convention, not a read rule, so this keys off the name and not the position. Confirmed against KiCad 10.0.5 — a sparse class at classes[0] with a complete Default behind it plots fine. The completeness requirement belongs to the Default alone. Named classes inherit every key they omit, so writing a full set for them would sever that inheritance and freeze them against later edits to the Default. They are untouched here. An existing incomplete Default cannot recover on its own, so an update backfills the keys that are absent and leaves every key that is present alone. Applying the creation defaults on update is mixelpixx#220, which is why they still do not apply here. Fixes mixelpixx#326.
Beyond mixelpixx#326, which only needed the Default written complete. Each of these stands alone and can be dropped without affecting that fix. Netclass priorities were hardcoded to 0. KiCad gives the Default std::numeric_limits<int>::max() and every other class -1 (net_settings.cpp:69, netclass.cpp:57), so a class this tool created sorted differently from one KiCad wrote. A written Default replaces KiCad's seeded one rather than merging with it, so the PCB values it carries become the project's. This tool's schema defaults are not KiCad's, which made create_netclass(name="Default") re-spec a board's routing rules as a side effect of a call meant to do something else. The Default now takes KiCad's values (netclass.cpp:36-50); the schema defaults still apply to named classes. get_netclasses reported a key absent from a class as null, which reads as "unset" when it means "inherits" — addMissingDefaults fills it from the Default at load. Settings are now reported resolved, with `inherits` naming what came from the Default. `missing_fields` on the Default names what nothing can resolve, which is the mixelpixx#326 state a caller otherwise has no way to see. Callers reading null as "unset" will need updating. create_netclass also reports `is_default`, and says what it repaired when it backfills.
mixelpixx
added a commit
that referenced
this pull request
Aug 29, 2026
Minor: three new tools and two response-shape changes. New tools: - set_predefined_sizes / get_predefined_sizes (verification, #346) — the PCB Editor's track/via palette, which is a third .kicad_pro key distinct from DRC floors and netclass optima. - delete_graphics (pcb_board, #335) — the missing generic delete verb, so a board outline can be replaced instead of only appended to. Response-shape changes, both toward reporting what is true: - get_netclasses reports settings resolved, with `inherits` naming what a class takes from the Default and `missing_fields` naming what nothing resolves (#333). Callers reading null as "unset" need updating: null claimed a class had no clearance when it inherited the Default's. - auto_place_from_schematic and refine_placement_force_directed report `held`, naming every footprint they refused to move and why (#350). Correctness: - Neither placement planner relocates a footprint KiCad has locked. The lock was never parsed at all, so both planners moved locked parts (#350). - update_footprints_from_library accepts the fp_text user that 15,238 of 15,447 official footprints carry, losslessly (#331). - design_review, sch_analysis, sch_export and sch_batch resolve pins through the placed unit; #182 is closed. - The Default netclass is written complete, and a incomplete one repaired (#326) — junction dots were suppressed project-wide. - Per-user Windows KiCad installs are discovered (#254); MCP startup no longer reinstalls guidance after an explicit uninstall (#242); placed symbols carry Datasheet and Description from the library (#226). Project: - GOVERNANCE.md states how the project is run; CODEOWNERS registers areas. - `cargo xtask fix-doc-counts` derives every documented count from the registry, so tool-adding PRs no longer conflict by construction. 20 toolsets, 217 registered tools, 223 total.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #326.
Fixing this turned up adjacent problems in the same code — netclass priorities, the Default's PCB values, how
get_netclassesreports an omitted key. Rather than fold them in, they're split by what they touch: the first commit changes only the class namedDefault, the second changes everything else.The first commit does stands alone, if you'd prefer a small pr.
fix(pcb)— the fixThe triage suggested extending the
FIELDStable, but that table drives creation for every class, and only the Default must be complete:addMissingDefaultsfills other classes from the Default, so a named class omitting a key is inheriting, not broken. Completing them would sever that. They're untouched here.Defaultreplaces KiCad's seeded one (net_settings.cpp:216) rather than merging. An update now backfills absent keys only, leaving present ones alone, so fix(netclass): stop create_netclass resetting values it was not givenFix/create netclass partial update #220 stays fixed.SetNamesetsm_isDefaulton an exact match (netclass.h:96) and the loader has no index branch. KiCad's writer emits the Default first, so it's alwaysclasses[0]in KiCad-written files — an output convention, not a read rule. Completing whichever class comes first would leave the bug live.refactor(pcb)— beyond the fixPriorities now match KiCad (
intmax for Default,-1otherwise). The Default takes KiCad's PCB values rather than this tool's schema defaults, since a written Default replaces the seeded one and was silently re-speccing boards — worth your explicit call, happy to split it out. Andget_netclassesreports settings resolved withinherits/missing_fieldsinstead ofnull, which breaks callers readingnullas "unset".Testing
Nine unit tests plus an
#[ignore]e2e test that drives the MCP tools and plots through eeschema. Against KiCad 10.0.5 it passes with the fix and fails without it. The junction dot's radius is the sharpest signal — 0.4572 mm healthy, 0.0001 mm when the Default is incomplete — so the test compares against a baseline plot rather than a literal. Note a barestroke:nonegrep false-positives: a healthy junction dot is filled, not stroked.