Let the template update a block a printer already has

.github/CONTRIBUTING.md

@@ -120,8 +120,10 @@ git checkout -B _sync2 origin/alpha && git merge --no-ff origin/beta && git push
   same change. Its config section arrives with the same update, and a section whose module is missing
   is not a missing feature — klippy refuses the whole config.
 - Changing `config-templates/AddOn.cfg.template`? Read the header of `scripts/addon_merge.py` first.
-  That file is never regenerated on a printer that already has one, so only whole `#@FEAT` blocks
-  reach existing machines. A change *inside* an existing block reaches nobody.
+  That file is never regenerated on a printer that already has one, so a new `#@FEAT` block is what
+  normally reaches existing machines. A change *inside* a block reaches nobody unless the template
+  also names that block in a `#@REVISE`, which replaces the printer's copy of it in place — read what
+  that costs before reaching for it.
 
 ## Version numbers
 

.github/ci/check-addon-markers.py

@@ -0,0 +1,123 @@
+#!/usr/bin/env python3
+"""Every marker in AddOn.cfg.template does something, or the build stops.
+
+addon_merge.py guards itself well at RUN time: an unpaired #@DROP is refused, a #@RETIRE for a
+feature the template still ships is ignored, a #@REVISE naming a block that is not there is skipped.
+Each of those is the right thing to do on a printer -- and each is silent to whoever wrote the
+marker, because the message is printed on a machine in somebody's workshop rather than in the pull
+request that got it wrong. The P114 gate ran two versions out of date for months for exactly this
+kind of reason: the delivery looked like it was working.
+
+So the same rules are checked here, where a typo is still cheap.
+"""
+import re
+import sys
+import os
+
+TPL = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "..",
+                   "config-templates", "AddOn.cfg.template")
+
+FEAT = re.compile(r'^#@FEAT\s+(\S+)\s*\|\s*(.*)$')
+RETIRE = re.compile(r'^#@RETIRE\s+(\S+)\s*\|\s*(.*)$')
+REVISE = re.compile(r'^#@REVISE\s+(\S+)\s*\|\s*(.*)$')
+DROP = re.compile(r'^#@DROP\s+(.+?)\s*\|\s*(.*)$')
+END = "#@ENDFEAT"
+OFF = "#:off:"
+SECTION = re.compile(r'^(?:' + re.escape(OFF) + r')?\[([^\]]+)\]\s*$')
+
+
+def main():
+    path = sys.argv[1] if len(sys.argv) > 1 else TPL
+    lines = open(path, encoding="utf-8").read().split("\n")
+    bad = []
+
+    # 1. Every block is closed, and closed before the next one opens. addon_merge refuses to touch an
+    #    unterminated block precisely because removing it would swallow the neighbour -- which in this
+    #    file means the two print macros.
+    feats, open_at, ids = {}, None, []
+    for n, l in enumerate(lines, 1):
+        m = FEAT.match(l)
+        if m:
+            if open_at is not None:
+                bad.append("line %d: #@FEAT %s opens while %s is still open (a missing #@ENDFEAT)"
+                           % (n, m.group(1), open_at[1]))
+            open_at = (n, m.group(1))
+            ids.append(m.group(1))
+            feats[m.group(1)] = [n, None]
+        elif l.startswith(END):
+            if open_at is None:
+                bad.append("line %d: #@ENDFEAT with no #@FEAT above it" % n)
+            else:
+                feats[open_at[1]][1] = n
+                open_at = None
+    if open_at is not None:
+        bad.append("line %d: #@FEAT %s is never closed" % (open_at[0], open_at[1]))
+
+    dupes = sorted({f for f in ids if ids.count(f) > 1})
+    for d in dupes:
+        bad.append("#@FEAT %s is declared more than once — the merge would only ever see the first" % d)
+
+    shipped = set(feats)
+
+    # 2. A #@RETIRE for a feature the template still ships is ignored at run time (addon_merge line
+    #    ~231). Silently: the block stays on every printer and the author believes it went.
+    for n, l in enumerate(lines, 1):
+        m = RETIRE.match(l)
+        if m and m.group(1) in shipped:
+            bad.append("line %d: #@RETIRE %s, but the template still ships that #@FEAT — "
+                       "the retirement would be ignored" % (n, m.group(1)))
+
+    # 3. A #@REVISE is the mirror image: it can only act on a block the template also ships, because
+    #    the replacement body comes from that block.
+    revised = set()
+    for n, l in enumerate(lines, 1):
+        m = REVISE.match(l)
+        if not m:
+            continue
+        revised.add(m.group(1))
+        if m.group(1) not in shipped:
+            bad.append("line %d: #@REVISE %s, but no #@FEAT %s in this template — "
+                       "there is nothing to revise it to" % (n, m.group(1), m.group(1)))
+    for f in sorted(revised & {m.group(1) for m in (RETIRE.match(l) for l in lines) if m}):
+        bad.append("#@REVISE and #@RETIRE both name %s — the merge would do neither" % f)
+
+    # 4. A #@DROP must be paired with a block that hands the section back, or it is a delete
+    #    instruction with nothing behind it.
+    in_block = set()
+    for fid, (a, b) in feats.items():
+        if b is None:
+            continue
+        for l in lines[a - 1:b]:
+            m = SECTION.match(l)
+            if m:
+                in_block.add(m.group(1).strip())
+    for n, l in enumerate(lines, 1):
+        m = DROP.match(l)
+        if m and m.group(1) not in in_block:
+            bad.append("line %d: #@DROP %s, but no #@FEAT block in this template declares that "
+                       "section — the merge would refuse to remove it" % (n, m.group(1)))
+
+    # 5. Two sections of the same name in one template is a config klippy refuses, and it would be
+    #    delivered to every fresh install.
+    seen = {}
+    for n, l in enumerate(lines, 1):
+        m = SECTION.match(l)
+        if not m:
+            continue
+        name = m.group(1).strip()
+        if name in seen:
+            bad.append("line %d: [%s] is declared twice (first at line %d)" % (n, name, seen[name]))
+        seen[name] = n
+
+    if bad:
+        print("AddOn.cfg.template markers:")
+        for b in bad:
+            print("  ERROR %s" % b)
+        return 1
+    print("AddOn.cfg.template: %d features, %d revisions, every marker resolves."
+          % (len(shipped), len(revised)))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

.github/workflows/ci.yml

@@ -196,6 +196,12 @@ jobs:
       - name: Anchors and step numbers
         run: python3 .github/ci/check-docs.py
 
+      # A marker that resolves to nothing is refused on the printer and silent to whoever wrote it:
+      # the message goes to a journal in somebody's workshop, not to the pull request. The P114 gate
+      # ran two versions out of date for months for that kind of reason. Same rules, checked here.
+      - name: AddOn.cfg markers
+        run: python3 .github/ci/check-addon-markers.py
+
       # The documents overlap on purpose and drift for free. This fails on text that has quietly
       # acquired a second owner, and -- when there is a base commit to compare against -- points out
       # sentences elsewhere that closely matched the wording just changed here. The second half is a

README.html

@@ -985,7 +985,7 @@ <h2 id="orcaslicer-multicolor-ams-auto-mode">OrcaSlicer — multicolor
 cell. <strong><code>G30</code></strong> then loads your saved
 <code>phrozen</code> mesh instantly — calibrating and saving it the
 first time, if none exists yet. Homing is automatic. It needs the
-<code>auto_mode2</code> and <code>ams</code> features in
+<code>auto_mode</code> and <code>ams</code> features in
 <code>AddOn.cfg</code>, both on by default, and the
 <code>gcode_shell_command</code> extension.</p>
 </details>

README.md

@@ -498,7 +498,7 @@ colour** — otherwise a long print changes shade partway through, and nothing w
 The start G-code heats the **bed straight to print temperature** and holds the **nozzle at a probe-safe
 140 °C** through home and probe, so it cannot ooze onto the load cell. **`G30`** then loads your saved
 `phrozen` mesh instantly — calibrating and saving it the first time, if none exists yet. Homing is
-automatic. It needs the `auto_mode2` and `ams` features in `AddOn.cfg`, both on by default, and the
+automatic. It needs the `auto_mode` and `ams` features in `AddOn.cfg`, both on by default, and the
 `gcode_shell_command` extension.
 
 </details>

config-templates/AddOn.cfg.template

@@ -773,14 +773,20 @@ gcode:
 #@RETIRE ams | AMS on/off switch - the AMS is detected now, so there is nothing to declare
 #
 
-# PHROZEN_TOOLCHANGE lives inside this block, and a change *inside* a block reaches fresh flashes
-# and nowhere else -- the merge adds and removes whole blocks, it never rewrites one a printer
-# already has. So the corrected macro ships under a new id and the old block is retired in the same
-# run, under the same backup, exactly as the g30 -> g30_meshdone replacement did. The pairing is
-# enforced by addon_merge, not trusted: if this block cannot be added, the retirement does not run,
-# so the two print macros this block carries can never go missing.
-#@RETIRE auto_mode | replaced by auto_mode2, which fixes a retract that assumed relative E
-#@FEAT auto_mode2 | Orca auto AMS-mode: reads the 'ams' flag -> P0 M1/M2/M3 at print start
+# PHROZEN_TOOLCHANGE lives inside this block. Blocks are the unit the merge moves, so an edit made
+# inside one used to reach fresh flashes and nowhere else -- which is why this marker exists. It says:
+# this block is the kit's, bring an existing printer's copy up to the version below. The owner's
+# on/off toggle is carried over, a block that already matches is left alone and not even mentioned,
+# and a block the owner deleted by hand is NOT put back -- that stays the seeded record's decision.
+# One printer generation carried this block as `auto_mode2`, from the short-lived attempt to deliver
+# the retract fix by renaming the block instead of revising it. #@REVISE cannot reach those: they have
+# no `auto_mode` to revise. They get `auto_mode` through the normal add path, and this line takes the
+# old copy out in the same run -- without it such a printer would end up with both, which is two
+# [gcode_macro PHROZEN_TOOLCHANGE] sections and a config klippy refuses.
+#@RETIRE auto_mode2 | folded back into auto_mode, which is the name it should never have lost
+
+#@REVISE auto_mode | the toolchange retract assumed relative E; see the M83 note in the macro
+#@FEAT auto_mode | Orca auto AMS-mode: reads the 'ams' flag -> P0 M1/M2/M3 at print start
 # ───── Flag-driven AMS mode for OrcaSlicer (Arco Unleashed) ─────
 # The 'ams' flag follows the AMS device node, kept in step by [arco_tool_gate] -- nobody sets
 # it, and there is nothing to remember to set. This macro reads it at print start and applies