You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit fd7013e
Browse filesBrowse the repository at this point in the historyBrowse files
docs: CHANGELOG 1.8.0 section with migration notes (#527)
## What
Adds the `## 1.8.0 — 2026-10-27` section (with migration notes) to `CHANGELOG.md`, above `## 1.7.0`. `CHANGELOG.md` is the only file changed (+131/-0). Documentation only: no `src/`, `include/`, `docker/`, `scripts/`, `tests/` or workflow file is touched.
## Why
The engine release body is generated from git-log subjects only and carries no migration notes; the release notes users read are this file (and the GitHub release body edited from it after publish). The notes must be on main before the release dispatch.
## Content
- Compatibility and migration: exact pairing with codegen 1.8.0, regenerate C++ and relink every strategy library, no public header change, the new codegen `color` input type, the changed harness file `docker/run_json.py`.
- Faster `run_json` report handling (#517) with the measured table against 1.6.3 and 1.7.0 and the statement that the gap to 1.6.3 is not closed.
- Scratch placement restated (#518); the unused storage building block (#515; its allocation-gate sites are unreviewed, no bound claimed); a known limit (memory with never-filled pending orders, "a fix is planned for a later release").
- CI and release process: every merged change since 1.7.0 not listed above (#505-#510, #513, #514, #516, #519, #520, #522, #469, #481, #504). Every line is taken from the merged commit body; none claims a behaviour change for a tests-only or CI-only change.
Not in this commit (not on main at base `641349a`): the release gate, tmpwatch and the TvJO fix. Each gets one line from its merged PR body as a follow-up commit on this branch before this PR merges; this PR is meant to merge after those three.
## Checks
- Route: docs-only, `lab verify --via ci-only` (no sweep).
- Read as a committed diff by an independent reviewer: GO. All 18 engine commits `v1.7.0..641349a` have a bullet, none extra.
- Numbers: the #517 timings and RSS are the measurements in the #517 body; the ratios in the text were recomputed (1.77/0.36 = 4.9, 8.26/1.35 = 6.1, 8.94/5.39 = 1.66; 4.61/0.36 = 12.8, 22.31/1.35 = 16.5, 22.99/5.39 = 4.27).
- The heading date `2026-10-27` is the planned tag date and is corrected before the tag if the date moves.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
paired with [pineforge-codegen 1.8.0](https://github.com/pineforge-4pass/pineforge-codegen-oss/releases/tag/v1.8.0).
12
+
A minor release. It speeds up the `run_json` wrapper (the wrapper only, not the
13
+
engine), adds one unused storage building block to the library, states where
14
+
scratch files are placed, and carries release-gate and CI changes. The C ABI
15
+
(`PF_ABI_VERSION` 4), the report keys and the run-failure code catalog are
16
+
unchanged (0 codes added, changed or removed since 1.7.0).
17
+
18
+
### Compatibility and migration
19
+
20
+
- Pairs with codegen 1.8.0 exactly. Regenerate strategy C++ with codegen 1.8.0
21
+
and relink every strategy library against the 1.8.0 headers and
22
+
`libpineforge.a`, as for every pair change; replacing the archive alone does
23
+
not update a statically linked strategy library.
24
+
- No public header changed. The only new header is
25
+
`include/pineforge/detail/o1_store.hpp`, an internal building block that no
26
+
engine path uses yet (below); its object code is part of `libpineforge.a`,
27
+
which therefore grows slightly.
28
+
- Hosts that read the codegen input manifest must recognize the new input type
29
+
`color` (a packed unsigned `0xAARRGGBB` integer default). That is a codegen
30
+
change, described in the codegen 1.8.0 notes; the engine side is unchanged:
31
+
color inputs are still stored and reported as `int64` through
32
+
`get_input_int64`.
33
+
- The release harness file `docker/run_json.py` changed (below). The hub image
34
+
for this pair carries the 1.8.0 copy.
35
+
36
+
### Faster `run_json` report handling
37
+
38
+
-**Fewer passes over the report sink (#517).**`docker/run_json.py`
39
+
now takes about three full passes over the report sink instead of about
40
+
eleven. The verify pass, which already reads the whole sink for its digest,
41
+
also folds the trade summary, the first and last equity points and the equity
42
+
finiteness check; the annualization block gets a two-point object instead of
43
+
five full re-reads (`point_at`); and the report writer emits flat records of
44
+
exact scalar types with one `json.dumps`, joined in batches fixed at 512
45
+
records and 256 KiB (a string field over 4096 characters keeps the windowed
46
+
path). The sink reader (`pf_sink.py`) is untouched. Trades and report
47
+
bytes are unchanged (identical under the accepted normalization of the timing
48
+
fields in all six measured cells) and peak RSS is unchanged (72.9 MB against
49
+
72.7 MB at 100k bars). Measured on one spot box, four repetitions per cell,
50
+
same tag build and shared library, only the wrapper replaced:
51
+
52
+
| bars | script | 1.6.3 | 1.7.0 | 1.8.0 |
53
+
|---|---|---|---|---|
54
+
| 20k | fixed ids | 0.36 s | 4.61 s | 1.77 s |
55
+
| 100k | fixed ids | 1.35 s | 22.31 s | 8.26 s |
56
+
| 100k | unique ids | 5.39 s | 22.99 s | 8.94 s |
57
+
58
+
This removes the wrapper's multi-pass cost, and it is a partial recovery: the
59
+
1.7.0 wrapper took about 4 to 17 times the 1.6.3 time on these cells (12.8x,
60
+
16.5x and 4.3x), and 1.8.0 takes about 5 times at 20k bars with fixed ids,
61
+
about 6 times at 100k bars with fixed ids and about 1.7 times at 100k bars
62
+
with unique ids. The gap to 1.6.3 is **not** closed. The engine's own
63
+
per-record sink parse and the dense-script close path are separate work, not in
64
+
this release.
65
+
66
+
### Scratch placement (behaviour of 1.7.0 and later, restated)
67
+
68
+
-**Where scratch goes and when a run is refused.** The text is in
69
+
[run-failure-codes](docs/pages/run-failure-codes.md) (#518) and restates the
70
+
1.7.0 rules for hosts that place scratch on memory-backed paths. Without an
71
+
explicit scratch directory, a native run with no scratch arena attached
72
+
creates one unlinked command-scratch file at begin, under the nonempty
73
+
`TMPDIR`, or under `/tmp` when `TMPDIR` is unset or empty. If that candidate
74
+
is proven tmpfs/ramfs, the engine falls back to `/var/tmp`, and only when
75
+
`/var/tmp` is disk-backed. If both candidates are memory-backed the run is
76
+
refused at begin: the engine raises `scratch_io_failed`, and the default
77
+
unjailed `run_json` profile refuses first with `run_request_invalid`;
78
+
through the hub entrypoint the refusal surfaces as container exit code 4.
79
+
An explicit `PINEFORGE_COMMAND_SCRATCH_DIR` on tmpfs/ramfs never falls back
80
+
and raises `scratch_io_failed`. Example: in a container image where `/run`
81
+
is tmpfs while `/tmp` and `/var/tmp` are on ext4, a work directory under
82
+
`/run` is memory-backed; a run with `TMPDIR` set there falls back to
83
+
`/var/tmp`. To choose the location, set `TMPDIR` to a disk-backed directory
84
+
such as `/var/tmp`; `PINEFORGE_COMMAND_SCRATCH_DIR` covers the command stores
85
+
only, not the trade ledger or the report files. This is not a change of
86
+
behaviour in 1.8.0.
87
+
88
+
### Storage building block
89
+
90
+
-**Fixed-shape O(1) store for Pine collections (#515).**
91
+
`pineforge::detail` gains a chunked store (`o1_store.hpp`, `o1_store.cpp`)
92
+
whose chunk buffer and resident pages are capped in the constructor
93
+
(`kMaxChunkBytes` 64 KiB, `kMaxResidentPages` 1024) with a typed refusal at
94
+
the cap, and transactions with rollback. Nothing in the engine uses it yet,
95
+
so no run changes. It is tested in `tests/test_o1_store.cpp`, and the
96
+
source allocation gate lists its new sites as unreviewed entries
97
+
(no bound is claimed for them).
98
+
99
+
### Known limits
100
+
101
+
-**Memory with never-filled pending orders.** On dense scripts,
102
+
memory still grows with the number of pending orders that never fill, by
103
+
about 2.5 kB each. A fix is planned for a later release.
104
+
105
+
### CI and release process
106
+
107
+
-**Light jobs on a pinned hosted image.** The light, secret-free CI jobs run
108
+
on a pinned GitHub-hosted image (#514); the AGENTS.md runner
109
+
routing is documented (#513).
110
+
-**Actions pinned to commit shas.** Every third-party GitHub Action is pinned
111
+
to a full commit sha with a version comment, and preflight lints it
112
+
(#508).
113
+
-**Release guard hardening.** The permission-claim text guard also catches
114
+
backslash-wrapped, bytes-decoded and fullwidth respellings (#506); the
115
+
release ruleset probe lists every branch ruleset and changes no gate verdict
116
+
(#505); the promote-baseline ancestry check and the doc-reverts default base
117
+
fully qualify their remote-tracking refs (#507).
118
+
-**Source allocation bound gate.** CI checks the engine and generated C++
119
+
against an allocation-bound inventory with scoped site identities
120
+
(#469). The inventory is not yet reviewed: its entries are unreviewed or
121
+
pending flat-memory work, and the gate fails on a new container site.
122
+
-**Release gate reads the post-merge run (#526).** The release gate
123
+
(`scripts/ci_release_gate.py`) now also requires a green, non-skipped
124
+
`ci-post` run of the commit being released, with all four build legs
125
+
(`ubuntu-24.04` and `macos-26`, Debug and Release) successful. Until now it
126
+
read only the `pineforge/verify` and `pineforge/parity` statuses. When the
127
+
release commit is documentation-only it checks the nearest earlier commit that
128
+
is not (at most 20 documentation-only commits), and every skipped
129
+
documentation-only commit needs a successful `ci-post` run of its own. A newer
130
+
run of the same commit with a failed, cancelled or timed-out build job
131
+
refuses the release; so does any other non-success result. A newer run whose build legs
132
+
were all skipped is passed over in favour of an older full green run. Runs on
133
+
other branches are not counted. The gate only reads results: it changes no
134
+
build or engine behaviour, and its command line is unchanged.
135
+
-**Live equivalence row.** Corpus equivalence runs as a required CTest row in
136
+
the live suites (#481).
137
+
-**Kernel residuals check.**`scripts/check_kernel_residuals.py` no longer
138
+
reports a free-standing `strings` line as an unruled identifier when no
139
+
symbol, installed header or decoded source literal corroborates it: the macOS
140
+
arm64 Release archive prints a four-byte machine-code run through
141
+
`strings -a`, which the check had read as an identifier. The limit is stated
142
+
in the change: a bare archive identifier with no corroborating evidence is no
143
+
longer detected by this check (#525, checker and its tests only, no engine
144
+
behaviour change).
145
+
-**Scratch-on-tmpfs tests ignore co-tenant files.** The default and jailed
146
+
scratch-on-tmpfs tests no longer fail when another job on the same CI machine
147
+
creates files in `/tmp`, `/var/tmp` or `/dev/shm`: the tmpfs candidate is a
148
+
private directory owned by the test, and in `/tmp` and `/var/tmp` only files
149
+
this process creates through `mkstemp` are counted. Tests only, no change to the
150
+
engine or its headers; the stated limit is that scratch created through
151
+
another call in those two roots would not be seen (#521).
152
+
-**Post-runner capability canary.** A dispatch-only workflow
153
+
(`fleet-post-capability.yml`) checks that a new post runner VM can run the
154
+
post jobs before it joins the post pool. It has no push, pull request or
155
+
schedule trigger, builds nothing and runs no engine suite (#520).
156
+
-**Tests and docs.** The `TIMEOUT 180` ctest property is dropped (#504); the
157
+
run-failure diff tests hold before and after a release reset (#509); the
158
+
doc-reverts test no longer inherits the caller's base (#516); release notes
159
+
for earlier engine releases are backfilled (#510); the three borrowed-fd
160
+
tests compare a descriptor's flags without Darwin's "descriptor was written"
161
+
bit, with diagnostics that name the flag and step that changed (#519, tests
162
+
only, no `src/` or `include/` change); the pytest-backed
163
+
`test_run_json_stream_fastpath` row no longer carries the `fast` label, so the
164
+
pull request release job's `fast` selection, which installs no pytest, does
165
+
not run it (#522, tests only).
166
+
8
167
## 1.7.0 — 2026-10-25
9
168
10
169
This release adds report-sink transport, scratch-backed history and bounded
0 commit comments