vm: prescriptive error for '+' on lists (#680)

CHANGELOG.md

@@ -102,6 +102,14 @@ All notable changes to EigenScript are documented here.
   max-subtraction (overflow to inf), a mis-indexed target, and sign errors.
 
 ### Changed
+- **`+` on two lists now names the fix (#680).** `[1,2] + [3,4]` reported
+  `cannot apply '+' to list and list` — correct but not actionable; a Python/JS
+  prior lands here and has to guess the idiom. The message now prescribes it:
+  `... (use 'append of [xs, v]' to add an element, or 'concat of [a, b]' to join
+  two lists)`, mirroring the existing f-string hint for mixed-type concatenation.
+  First of the prescriptive-error pass (#680) — errors at points where the
+  language diverges from mainstream priors should teach the fix, for humans and
+  AI alike. Regression: `test_error_extra.eigs` EE20b.
 - **The observer learning-rate gate is now two-channel (entropy *and* drift).**
   `observer_matrix_scale` throttled a weight matrix to 0.5x whenever its
   entropy was quiet — but entropy is computed over one SGD step's worth of

CLAUDE.md

@@ -18,6 +18,16 @@ session: a gap you route upstream is a contribution; a gap you paper over is
 lost data. Do this without being told, the way you'd run the tests without being
 told.
 
+## Before merging your own PR, clear ready contributor PRs first
+
+External contributors work from **forks we cannot push to** — a fork branch is
+theirs. If our merge makes their PR stale, the only remedies are asking them to
+rebase (friction, delay) or the fetch-rebase-reland dance (see #657). Our OWN PRs
+we rebase for free. So before merging one of ours: `gh pr list` and land any
+**ready** contributor PR (green + reviewed, not authored by us) first, then rebase
+ours onto the new main. Rebasing our own costs nothing; asking them does. (Only
+*ready* PRs — never merge an unready contributor PR just to dodge a rebase.)
+
 ## Build & test
 
 ```

src/vm.c

@@ -2481,6 +2481,12 @@ static Value *vm_run_ex(EigsChunk *chunk, Env *env, Task *resume) {
                 rt_error(EK_TYPE, current_line,
                     "cannot apply '+' to %s and %s (use an f-string or 'str of' to concatenate)",
                     ta, tb);
+            else if (a->type == VAL_LIST || b->type == VAL_LIST)
+                /* #680: name the fix, don't just diagnose. '+' never joins lists
+                 * (it is numbers-or-strings only); a Python/JS prior lands here. */
+                rt_error(EK_TYPE, current_line,
+                    "cannot apply '+' to %s and %s (use 'append of [xs, v]' to add an element, or 'concat of [a, b]' to join two lists)",
+                    ta, tb);
             else
                 rt_error(EK_TYPE, current_line, "cannot apply '+' to %s and %s", ta, tb);
             vm_push(make_null());

tests/test_error_extra.eigs

@@ -227,6 +227,18 @@ catch e:
 assert of [caught20 == 1, "EE20 caught '+' on dict and num"]
 assert of [(contains of [msg20.message, "cannot apply '+'"]) == 1, "EE20 message mentions cannot apply '+'"]
 
+# ---- '+' on two lists NAMES THE FIX (#680: prescriptive, not just diagnostic) ----
+caught20b is 0
+msg20b is ""
+try:
+    r is [1, 2] + [3, 4]
+catch e:
+    caught20b is 1
+    msg20b is e
+assert of [caught20b == 1, "EE20b caught '+' on list and list"]
+assert of [(contains of [msg20b.message, "append of"]) == 1, "EE20b list '+' message prescribes append"]
+assert of [(contains of [msg20b.message, "concat of"]) == 1, "EE20b list '+' message prescribes concat"]
+
 # ---- '/' on incompatible types ----
 caught21 is 0
 msg21 is ""