Stream pull request diffs to disk, and prune CI's uploaded runs []

.agents/skills/edg-python-style/SKILL.md

@@ -26,7 +26,8 @@ Prefer matching the file you are editing over inventing a new dialect.
   neighbors.
 - Package code lives under `dev_tools/pylibs/<pkg>/` (importable as
   `edgtest`, `edgutil`, `edgacknowledg`, …). Put `dev_tools/bin` on `PATH`
-  and `dev_tools/pylibs` on `PYTHONPATH` (direnv / [HACKING.md](../../../HACKING.md)).
+  and `dev_tools/pylibs` on `PYTHONPATH` (direnv /
+  [HACKING.md](../../../HACKING.md)).
 
 ## Indentation and wrapping
 
@@ -36,41 +37,143 @@ Prefer matching the file you are editing over inventing a new dialect.
 - Wrap long signatures, calls, and `typing` imports with parentheses; align
   continuations with neighboring EDG Python, not Black/ruff defaults.
 
-### Function parameter lists
+### Function calls and wrapped message text
+
+Call-site wrapping differs from the declaration wrapping below. When a
+call's arguments need multiple lines:
+
+- Break immediately after the opening `(`; put the arguments on following
+  lines, indented **two spaces** from the statement's indentation.
+- Once a call is split over multiple lines, put **each positional and
+  keyword argument on its own line**, even when several would fit together.
+  A single argument may span multiple lines (for example, a long message),
+  but do not put another argument on any of those lines.
+- Put the closing `)` on its own line, aligned with the statement's start
+  (including `raise`, `return`, or `await`). Keep suffixes such as `from
+  None` or a `with` statement's `:` after that closing parenthesis.
+- Do not leave the first argument beside the opening parenthesis and hang
+  subsequent arguments under it. Declaration-style right alignment does
+  not apply to calls.
+
+After moving arguments onto new lines, **reflow their text for the new
+available width**. Count indentation, quotes, and any `f` prefix against
+79 columns. Merge adjacent message literals when the whole argument fits
+on one line; otherwise split at word boundaries using the available width.
+Do not preserve old string breaks just because they belonged to the old
+hanging layout. Preserve exact spaces, interpolation, and rendered text;
+implicit concatenation must not introduce a newline into the message.
+
+For example, replace this hanging call:
 
-If a `def` / `async def` signature does not fit on one line, wrap the
-parameter list in the same spirit as EDG C++ (`code_style.rst` — Wrapping
-Function Declarations):
+```python
+raise ConfigError(f"{host_key} must be a host name, without a scheme, "
+                  f"port, or path")
+```
 
-1. Prefer keeping the `def` name and opening `(` on the first line when the
-   first parameter(s) still fit; continue parameters on following lines.
-2. When that is not enough (long name, many parameters, long annotations),
-   break immediately after `(` and put parameters on the next lines.
-3. **Right-indent** wrapped parameter lines so they end at **column 79**
-   (right-aligned to the margin), packing as many parameters per line as
-   fit. Do not use a fixed left hanging indent (PEP 8 / Black’s 4-space
-   continuation).
+with this layout, whose message now fits on one argument line:
+
+```python
+raise ConfigError(
+  f"{host_key} must be a host name, without a scheme, port, or path"
+)
+```
+
+When the message still needs wrapping, reflow the words inside the call:
+
+```python
+raise ConfigError(
+  f"{port_key} is required for a host other than localhost "
+  f"({TLS_PORT} for TLS)"
+)
+```
+
+For a call with multiple arguments, keep each argument separate:
+
+```python
+return replace(
+  self,
+  service_bind = service_bind,
+  builder_service_url = builder_service_url,
+  publisher_service_url = publisher_service_url
+)
+```
+
+Review argument separation, parentheses, and text wrapping in added or
+modified calls. An unchanged syntax tree verifies behavior, not readable
+layout.
+
+### Multiline collections
+
+Optimize for readability and small future diffs, not fewer lines. In a
+multiline list, tuple, set, or dictionary literal, put **one element or
+key-value pair per line**, even when several would fit. Keep trailing
+commas so adding or removing an element changes only its own line. This
+also applies to collections passed to calls such as `frozenset`.
+Naturally short single-line collections can remain on one line.
+
+```python
+_INT_KEYS = frozenset((
+  'max_artifact_bytes',
+  'max_extracted_bytes',
+  'max_artifact_entries',
+))
+```
+
+### Function parameter lists
+
+For `def` / `async def` signatures, adapt the **Wrapping Function
+Declarations** rules in
+[the EDG C++ style guide](../../../doc/source/code_style.rst):
+
+1. First try wrapping with continuation parameters **left-aligned just to
+   the right of the opening `(`**, in the first parameter's column. Keep
+   initial parameters on the declaration line when they fit.
+2. Use that layout only if every resulting line fits within **79 columns**.
+   If the declaration overflows and a parameter cannot fit at that column,
+   break immediately after the opening `(` and move all parameters below it.
+3. In this fallback layout, use a **common indentation column chosen by
+   right-aligning the longest continuation line to column 79**. Shorter
+   lines start in that same column; do not right-justify each line
+   independently. Include closing punctuation and any return annotation
+   when measuring the longest line.
+4. Treat the keyword-only separator `*,` as belonging to the preceding
+   parameter when wrapping. Keep it on that parameter's line whenever it
+   fits within 79 columns, **including the first declaration line**:
+   write `def with_wiring(self, *,` rather than putting `*,` below `self`.
+   The same applies after `cls` or another positional parameter, even when
+   all following keyword-only parameters wrap. Do not give `*,` its own
+   line when the preceding line has room for it.
+
+When reviewing a wrapped signature, check lines beginning with `*,` and
+move the separator onto the preceding line if it fits.
+
+Normal left alignment (including `ForgeStore.open`):
 
 ```python
 def get_output_filename(env: Dict[str, str],
                         case_number: int,
                         command_number: int) -> str:
 
+async def open(cls, config: DatabaseConfig, *,
+               password: Optional[str] = None) -> 'ForgeStore':
+
+def with_wiring(self, *,
+                service_bind: Optional[str] = None) -> 'ForgeServiceConfig':
+```
+
+Fallback when left alignment would overflow:
+
+```python
 async def invoke_recorded_process(
-                    base_command: List[str], command_args: List[str],
-                    working_dir: Path, env: Dict[str, str],
-                    build_constants: BuildConstants,
-                    output_file_handle, *,
-                    filter_command: Optional[str] = None, debug: bool,
-                    driver_debug: bool = False,
-                    injected_command_args: List[str] = None) -> int:
-
-def parse_test_status_diff(
-                         status_line: str) -> Tuple[str, ParsedTestStatusDiff]:
+                     base_command: List[str],
+                     command_args: List[str],
+                     working_dir: Path,
+                     injected_command_args: Optional[List[str]] = None) -> int:
 ```
 
 Return annotations (`-> …`) stay on the last parameter line when they fit
-within 79 columns; otherwise wrap similarly to neighbors.
+within 79 columns; otherwise wrap similarly to neighbors. These declaration
+rules do not change the call-site rules above.
 
 ## Spacing around `=`
 
@@ -127,6 +230,14 @@ Typical order:
 Avoid unused imports; do not introduce heavy formatters that rewrite the
 whole tree.
 
+## Count checks
+
+When a condition tests whether a collection's element count is zero or
+nonzero, prefer explicit `len(items) == 0` or `len(items) != 0` over
+collection truthiness (`if not items:` or `if items:`). For example, write
+`if len(removed) != 0:` when deciding whether there are removed settings to
+report. This rule concerns counts, not boolean flags or optional values.
+
 ## Structure
 
 - Modules are flat and explicit; classes are fine for stateful helpers.
@@ -139,9 +250,10 @@ whole tree.
 ## What not to do
 
 - Do not run Black/ruff/autopep8 in a way that converts the tree to 4-space
-  PEP 8, strips spaces around `=`, or replaces right-aligned-to-79
-  parameter wraps with fixed hanging indents.
-- Do not copy front-end **C** style (`edg-cpfe-style`) into Python.
+  PEP 8, strips spaces around `=`, or replaces the two declaration-wrapping
+  layouts above with fixed four-space continuations.
+- Do not copy unrelated front-end **C** syntax/layout conventions into
+  Python; use the declaration-wrapping adaptation above.
 - Do not invent a second string-quoting scheme; follow `f"..."` / `'...'`.
 
 ## Related skills

.github/workflows/acknowledg-forge-deploy.yml

@@ -13,17 +13,15 @@
 #            installed copy of
 #            dev_tools/services/acknowledg-forge/deploy-remote.py, the only
 #            thing the deploy user may run as root) to point the service at
-#            the pushed digest, set its AcknowlEDG API and UI URLs, and
-#            restart it (only if any of those changed)
+#            the pushed digest and restart it when that changes
 #
 # The image is public and holds no credentials or configuration: the publish
 # job has no access to the deployment's secrets. Everything else about the
 # service (its unit, tokens, API key) is configured on the host.
 #
 # Host, user, and SSH credentials come from secrets in the
-# `acknowledg-forge` environment; the public AcknowlEDG URLs come from
-# repository variables shared with acknowledg-deploy.yml (see
-# dev_tools/services/acknowledg-forge/README.md).
+# `acknowledg-forge` environment. Application settings, including public
+# AcknowlEDG endpoints, belong in the host administrator's TOML.
 
 name: acknowledg-forge Deploy
 
@@ -130,11 +128,8 @@ jobs:
         env:
           SERVICE: ${{ secrets.ACKNOWLEDG_FORGE_SERVICE || 'acknowledg-forge-github' }}
           IMAGE: ${{ needs.publish.outputs.image }}
-          API_URL: ${{ vars.ACKNOWLEDG_API_HOST }}
-          UI_URL: ${{ vars.ACKNOWLEDG_UI_HOST }}
         run: |
           # The arguments are deliberately expanded (and quoted) here.
           # shellcheck disable=SC2029
           ssh acknowledg-forge "sudo -n /usr/local/libexec/acknowledg-forge-deploy \
-            $(printf '%q ' "${SERVICE}" "${IMAGE}" "${API_URL}" \
-              "${UI_URL}")"
+            $(printf '%q ' "${SERVICE}" "${IMAGE}")"

.github/workflows/build-and-test-ci.yml

@@ -308,7 +308,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     container:
-      image: ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     defaults:
       run:
         shell: edg-container-exec "bash -eo pipefail {0}"

.github/workflows/ci-toolchain.yml

@@ -66,7 +66,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     container:
-      image: &ci-runner-image ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: &ci-runner-image ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     defaults: &container-defaults
       run:
         shell: edg-container-exec "bash -eo pipefail {0}"
@@ -267,6 +267,17 @@ jobs:
 
           edg-prefix-env-loader .tmp/extensive-ci.edgivk
 
+      # Reviews (acknowledg-forge, edgy-review) read only the .rt.tar.gz
+      # files changes.elist names; edgy -A archives every test's. Pruning
+      # must never keep the results from being uploaded, hence
+      # continue-on-error.
+      - name: Prune .rt.tar.gz results changes.elist doesn't name
+        if: ${{ failure() }}
+        continue-on-error: true
+        env:
+          RUN_TIMESTAMP: ${{ inputs.run_timestamp }}
+        run: edg-prune-test-run "tests/runs/${RUN_TIMESTAMP}"
+
       - name: Upload partial elist, elog, and .rt.tar.gz results
         if: ${{ failure() }}
         uses: actions/upload-artifact@v7

.github/workflows/promote-test-cache.yml

@@ -72,7 +72,7 @@ jobs:
     # The test legs' image, so that actions/cache picks the same compression
     # (part of an entry's version) and the test legs can restore the entry.
     container:
-      image: ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     strategy:
       fail-fast: false
       matrix:

.github/workflows/strict-build-check.yml

@@ -158,7 +158,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     container:
-      image: ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     defaults:
       run:
         shell: edg-container-exec "bash -eo pipefail {0}"
@@ -185,7 +185,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     container:
-      image: ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     defaults:
       run:
         shell: edg-container-exec "bash -eo pipefail {0}"
@@ -212,7 +212,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     container:
-      image: ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     defaults:
       run:
         shell: edg-container-exec "bash -eo pipefail {0}"
@@ -239,7 +239,7 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     container:
-      image: ghcr.io/edgcpp/ci-runner@sha256:565d7b6581912008ca91c5dbc940280c6e28a02c8cb6487d88ee5f0d7e8c51d7
+      image: ghcr.io/edgcpp/ci-runner@sha256:daae67adf04085b9ae54a816b35053b79086070434d2da4aafd6662f6ecb0bf7
     defaults:
       run:
         shell: edg-container-exec "bash -eo pipefail {0}"

dev_tools/bin/edg-prune-test-run

@@ -0,0 +1,58 @@
+#!/usr/bin/env python3
+
+# Part of the EDG Compiler Project, under the Apache License v2.0 with LLVM
+# Exceptions.
+# See https://edgcpp.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+'''Prune a test run down to what a review of it reads, e.g., before CI
+uploads the run:
+
+  edg-prune-test-run tests/runs/<run timestamp>
+
+Reviews of a run read only the .rt.tar files its changes.elist files name.
+The replays directory is removed whole.  In each configuration, a suite
+with no test named there is removed whole (its directory, and its
+suite-<name> logs), as is every suite of a configuration without a
+changes.elist; of the other suites, the .rt.tar files not named there are
+removed.
+'''
+
+import argparse
+import sys
+
+import edgtest
+import edgutil
+
+from pathlib import Path
+
+def _parse_args() -> argparse.Namespace:
+  parser = argparse.ArgumentParser(
+    description = (
+      "Remove the .rt.tar files of a test run that no configuration's "
+      'changes.elist names.'
+    )
+  )
+  parser.add_argument(
+    'run_dir',
+    type = Path,
+    help = 'the run directory (e.g., tests/runs/<run timestamp>)'
+  )
+  return parser.parse_args()
+
+def main() -> int:
+  args = _parse_args()
+  if not args.run_dir.is_dir():
+    edgutil.eprint(f"error: {args.run_dir} is not a directory")
+    return 1
+
+  # Counts: of .rt.tar files kept and removed, and of suites removed.
+  kept, removed, suites = edgtest.prune_run_archives(args.run_dir)
+  print(
+    f"Kept {kept} .rt.tar files named by changes.elist files; removed "
+    f"{removed} others, and {suites} suites with no changes."
+  )
+  return 0
+
+if __name__ == '__main__':
+  sys.exit(main())