The suites import std/test, and the fourth harness copy is gone

.github/workflows/ci.yml

@@ -158,10 +158,10 @@ jobs:
       # would have. Pinned rather than floating: a green run should mean this
       # code passed against a known compiler, not against whatever was newest
       # that morning.
-      - name: Install twill v1.9.0
+      - name: Install twill v1.12.0
         run: |
           curl -fsSL -o twill \
-            https://github.com/twill-lang/twill/releases/download/v1.9.0/twill-v1.9.0-linux-amd64
+            https://github.com/twill-lang/twill/releases/download/v1.12.0/twill-v1.12.0-linux-amd64
           chmod +x twill
           ./twill --version
 

CHANGELOG.md

@@ -2,6 +2,15 @@
 
 ## Unreleased
 
+### Changed
+
+- **The suites are written with `std/test`.** `tests/harness.tw` is deleted.
+  It was the fourth copy of the same harness across four repositories, and
+  `docs/needs.md` entry 12 said a `std/test` was what would delete it. twill
+  1.11 shipped one, so every `*_test.tw` imports `"std/test"` and calls the
+  same four assertions by the same names. The seven suites pass on twill
+  1.12.0, and the pin moved from 1.9.0 to 1.12.0 in `spool.toml` and CI.
+
 ### Added
 
 - **Warmup times itself.** Every pass is measured with `mono_ns`, and the report

README.md

@@ -28,14 +28,15 @@ the 6 test suites under `tests/` pass, the example loads a published model and
 answers with it, and CI runs both against a released twill on every push rather
 than gating on the prose in this file.
 
-You need twill 1.9.0 or newer. Get one:
+You need twill 1.11.0 or newer, because the suites are written with `std/test`,
+which arrived in 1.11. Get one:
 
 ```bash
-curl -fsSL -o twill https://github.com/twill-lang/twill/releases/download/v1.9.0/twill-v1.9.0-linux-amd64
+curl -fsSL -o twill https://github.com/twill-lang/twill/releases/download/v1.12.0/twill-v1.12.0-linux-amd64
 chmod +x twill
 ```
 
-The asset name is `twill-v1.9.0-<os>-<arch>`: `linux-amd64`, `linux-arm64`,
+The asset name is `twill-v1.12.0-<os>-<arch>`: `linux-amd64`, `linux-arm64`,
 `darwin-amd64`, `darwin-arm64`, `windows-amd64.exe`.
 
 The suite needs a checkout of [selvedge](https://github.com/twill-lang/selvedge)

docs/needs.md

@@ -283,15 +283,20 @@ line, now lit from the shared palette so it never drifts in colour.
 ### 12. A test runner
 
 **Would improve:** `tests/`
-**Status:** DELIVERED. `twill test tests` collects `*_test.tw`, runs each in a
-fresh interpreter and reports once. CI calls it and so does the README.
-`tests/harness.tw` stays, because the runner names the file that failed and the
-harness names the assertion inside it; deleting the three copies across three
-repositories wants a `std/test`.
-
-`tests/harness.tw` is now the fourth identical copy of the same file across four
-repositories. A `twill test` that collected `*_test.tw`, ran each in a fresh
-interpreter and reported once would delete all four.
+**Status:** DELIVERED, both halves. `twill test tests` collects `*_test.tw`,
+runs each in a fresh interpreter and reports once. CI calls it and so does the
+README. twill 1.11 then shipped `std/test`, which is what this entry said would
+delete the harness copies, and `tests/harness.tw` is deleted: every suite
+imports `"std/test"` and calls `check`, `equal_str`, `equal_i64` and `near` by
+the same names with the same signatures, so no assertion changed. `near` still
+takes the tolerance and never defaults it. The summary is now printed in the
+shape the runner reads, `<suite> passed <p> failed <f>` and then `OK` or
+`FAILED`, which the copy never was.
+
+What it said while the copy was here: `tests/harness.tw` stays, because the
+runner names the file that failed and the harness names the assertion inside
+it; deleting the three copies across three repositories wants a `std/test`. It
+was the fourth identical copy of the same file across four repositories.
 
 ### 13. Concurrency, and a process interface
 

spool.toml

@@ -11,7 +11,7 @@ entry = "src/predict.tw"
 # The language itself, pinned to the release whose std modules and tensor
 # builtins this source was written against. spool has no notion of a toolchain
 # dependency, so it is declared as an ordinary git dependency.
-twill = { version = "^1.9.0", git = "https://github.com/twill-lang/twill" }
+twill = { version = "^1.12.0", git = "https://github.com/twill-lang/twill" }
 
 # Reading model archives. A real dependency and not a convenience: loading
 # through selvedge gets the model's declared input and output shapes out of the

tests/batcher_test.tw

@@ -1,6 +1,6 @@
 mode systems
 
-import "harness.tw" as t
+import "std/test" as t
 import "../src/signature.tw" as sg
 import "../src/model.tw" as md
 import "../src/batcher.tw" as bat

tests/model_test.tw

@@ -12,7 +12,7 @@ mode systems
 # it back through shuttle. It is the only test in the repository that touches
 # the filesystem, and it cleans up after itself.
 
-import "harness.tw" as t
+import "std/test" as t
 import "../src/model.tw" as md
 import "../src/signature.tw" as sg
 import "../../selvedge/src/archive.tw" as arc

tests/predict_test.tw

@@ -1,6 +1,6 @@
 mode systems
 
-import "harness.tw" as t
+import "std/test" as t
 import "../src/signature.tw" as sg
 import "../src/model.tw" as md
 import "../src/predict.tw" as pr

tests/quant_test.tw

@@ -9,7 +9,7 @@ mode systems
 # unbiased, that the sizes are what the projection says, and that a saturating
 # calibration says how much it saturated.
 
-import "harness.tw" as t
+import "std/test" as t
 import "../src/signature.tw" as sg
 import "../src/model.tw" as md
 import "../src/quant.tw" as q

tests/score_test.tw

@@ -1,6 +1,6 @@
 mode systems
 
-import "harness.tw" as t
+import "std/test" as t
 import "../src/signature.tw" as sg
 import "../src/model.tw" as md
 import "../src/score.tw" as sc

tests/signature_test.tw

@@ -7,7 +7,7 @@ mode systems
 # debug, which is the situation twill's checker exists to prevent and which
 # serving otherwise reintroduces.
 
-import "harness.tw" as t
+import "std/test" as t
 import "../src/signature.tw" as sg
 import "std/text" as tx
 

tests/warmup_test.tw

@@ -9,7 +9,7 @@ mode systems
 # saving is only claimed when there is enough evidence for one, that a first
 # pass no slower than the rest claims nothing, and that every pass is timed.
 
-import "harness.tw" as t
+import "std/test" as t
 import "std/text" as tx
 import "../src/signature.tw" as sg
 import "../src/model.tw" as md