ci: push runs only the baseline; matrix on PR/manual; timeouts
ci / cell 16 (push) Skipped
ci / cell 32 (push) Skipped
ci / cell 8 (push) Skipped
ci / wide sweep (manual) (11) (push) Skipped
ci / wide sweep (manual) (16) (push) Skipped
ci / wide sweep (manual) (24) (push) Skipped
ci / wide sweep (manual) (32) (push) Skipped
ci / wide sweep (manual) (8) (push) Skipped
ci / cold cache (manual) (push) Skipped
ci / baseline (11 bit) (push) Canceled after 0s

One runner means one job at a time.  Run #1 took 30 minutes because a push
queued four jobs, each a cold build, and every later push waited behind them.
Now a push runs the 11 bit baseline alone (and not at all for docs only
changes), the width matrix waits for a pull request or the manual button, and
every job has a timeout so a stuck step fails instead of blocking the queue.
This commit is contained in:
dsh committed 2026-11-10 19:42:13 +08:00
1 parent 692995af59
commit a891538c9d
2 files changed
+131 -21

No files matched your search

+27 -21
View File
@@ -1,38 +1,44 @@
# Gitea act_runner. Labels: change `runs-on` to whatever label your runner
# advertises (a self hosted act_runner usually shows `ubuntu-latest` unless you
# gave it custom labels).
# Gitea act_runner (labels: ubuntu-latest / ubuntu-24.04 / ubuntu-22.04).
#
# push / PR : the 11 bit baseline (ctest, ~10 s) + the width matrix.
# manual : the same, plus the wide width sweep and a cold cache run (that one
# re-searches every synthesised block, so it is minutes not seconds).
# One runner means one job at a time, so the cheap tier has to stay cheap:
# push : baseline only, 11 bits, and not for docs-only commits
# PR : baseline + the width matrix
# manual : everything, including the wide sweep and the cold cache audit
# Every job has a timeout: a hung step used to sit in the queue for half an hour
# and block every later push.
name: ci
on:
push:
paths-ignore:
- "**.md"
- "docs/**"
- ".r8forge_cache/**"
pull_request:
workflow_dispatch:
jobs:
baseline:
name: baseline (11 bit)
runs-on: ubuntu-latest
timeout-minutes: 20
steps:
- uses: actions/checkout@v4
- name: configure
run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
- name: build
run: cmake --build build -j"$(nproc)"
run: cmake --build build -j 4
- name: ctest
run: ctest --test-dir build -j"$(nproc)" --output-on-failure
run: ctest --test-dir build -j 4 --output-on-failure
widths:
# A build of another cell width is one configure away (-DR8FORGE_CELL_SIZE).
# Only the `widthfree` tests are required here: examples/ pins 2047 style
# values and byte_op_test drives the 8/11 bit word macros, so neither means
# anything at 7 or 32 bits.
# Another cell width is one configure away (-DR8FORGE_CELL_SIZE). Only the
# `widthfree` tests are required: examples/ pins 2047 style values and
# byte_op_test drives the 8/11 bit word macros.
name: cell ${{ matrix.cell }}
if: github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch'
runs-on: ubuntu-latest
timeout-minutes: 30
strategy:
fail-fast: false
matrix:
@@ -42,14 +48,15 @@ jobs:
- name: configure
run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DR8FORGE_CELL_SIZE=${{ matrix.cell }}
- name: build
run: cmake --build build -j"$(nproc)"
run: cmake --build build -j 4
- name: width free tests
run: ctest --test-dir build -L widthfree -j"$(nproc)" --output-on-failure
run: ctest --test-dir build -L widthfree -j 4 --output-on-failure
sweep:
name: wide sweep (manual)
if: github.event_name == 'workflow_dispatch'
runs-on: ubuntu-latest
timeout-minutes: 60
strategy:
fail-fast: false
matrix:
@@ -59,9 +66,9 @@ jobs:
- name: configure
run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DR8FORGE_CELL_SIZE=${{ matrix.cell }}
- name: build
run: cmake --build build -j"$(nproc)"
run: cmake --build build -j 4
- name: width free tests
run: ctest --test-dir build -L widthfree -j"$(nproc)" --output-on-failure
run: ctest --test-dir build -L widthfree -j 4 --output-on-failure
cold-cache:
# The committed .r8forge_cache is trusted by every build, so on demand we
@@ -70,15 +77,14 @@ jobs:
name: cold cache (manual)
if: github.event_name == 'workflow_dispatch'
runs-on: ubuntu-latest
timeout-minutes: 60
steps:
- uses: actions/checkout@v4
- name: configure
run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
- name: build
run: cmake --build build -j"$(nproc)"
run: cmake --build build -j 4
- name: forget the cache
run: rm -rf .r8forge_cache
- name: search again
run: ctest --test-dir build -R 'flagship|synth_full_add' --output-on-failure
- name: the fresh results are what we ship
run: git status --short .r8forge_cache && test -d .r8forge_cache
run: ctest --test-dir build -R "flagship|synth_full_add" --output-on-failure
+104
View File
@@ -0,0 +1,104 @@
; calc_lex —— 桌上计算器 stage 1:只做字符分派(docs/calc.md §3 第 1 步)。
;
; 固定读 8 个字符(%rep 8,没有 EOF 处理),每读一个打一个数:
; '0'..'9'(0x30..0x39) -> 1 + digit (1..10)
; '*' '+' ',' '-' '.' '/'(0x2A..0x2F)-> 100 + (C - 0x2A) (100..105)
; 其它(空格、CR、字母……) -> 200
;
; 判法(bit4 是便宜的粗筛):
; * 字母的大写/小写都有 bit6=1,而数字和六个命令都没有 —— 先把 bit6 转到 bit0,
; 有就 erase C,让后面两个测试都跳过,R 留 200;
; * 再把 bit4 转到 bit0:数字 0x30..0x39 有 bit4=1,六个命令没有 —— biz/rnz 判 cell&1;
; * 剩下的 C 做一次减法 t = C - 0x2A(clb_sub_nc,常数 42 用 load 装):
; 命令正好 0..5,进位输出 Ci = 1 表示"没有借位"(C >= 0x2A),
; 空格 0x20 / CR 0x0D 都 < 0x2A,Ci = 0,R 留 200。
;
; 分支形状:每个条件体的身体都回到自己的条件格(净位移 0,见 AGENTS.md
; "语义上会咬人的几条");每轮先把 R 重置成 200,数字/命令分支只负责覆盖 R,
; 最后统一 output R —— 于是"跳过另一条分支"不需要 if/else。
;
; 验证(8 个字符,正好读满;六个命令和空格都在里面):
; build/Release/r8forge_derive.exe drv/calc_lex.r8asm 0 -i "9 *+,-./"
; in 9 *+,-./
; out 10 200 100 101 102 103 104 105
%include "asm/lib.r8asm"
%include "asm/raw_sub_nc.r8asm"
%include "asm/raw_add_nc.r8asm"
; Macro load
; %1: Cell operand
; %2: Count value
;
; 先把整格清干净,再 `.xor` 上常数(`.xor` 是异或,不清格就变成和旧值异或)。
%macro load 2
erase %1
.tp %1
.xor %2
%endmacro
.var C 0 ; 当前字符
.var B 1 ; load 的常数格 / 运算的 B 操作数
.var T 2 ; t = C - 0x2A(命令索引)
.var R 3 ; 这一轮要输出的数(默认 200)
.var Ci 4 ; 进位/借位
%rep 8
load R 200
.tp C
inp
; bit6 -> bit0:字母(含 0x50..0x5A 这种 bit4 也为 1 的大写)先出局
%rep 6
ror
%endrep
.tp C
biz
erase C ; 字母/其它:R 保持 200;C 清 0,后面两个测试自然跳过
rnz
; 恢复原字符(4 + 6 + 1 = 11 次旋转是一整圈)
%rep 5
ror
%endrep
; bit4 -> bit0:数字
%rep 4
ror
%endrep
.tp C
biz
; 数字:恢复原字符后 R = C - 47 = 1 + digit
%rep 7
ror
%endrep
load B 47
.tp Ci
set_1 ; Ci=1 -> clb_sub_nc 算 C - B
clb_sub_nc C B R Ci
.tp C ; 身体回到条件格;C 已被减法吃成 0
rnz
; 非数字:恢复原字符(数字分支吃掉的 C=0 转多少次还是 0)
%rep 7
ror
%endrep
; 命令索引 t = C - 0x2A;Ci = 1 表示 C >= 0x2A
load B 42
.tp Ci
set_1
clb_sub_nc C B T Ci
.tp Ci
biz
; 命令:R = t + 100
load B 100
.tp Ci
set_0 ; Ci=0 -> 普通加法
clb_add_nc T B R Ci
.tp Ci ; 身体回到条件格;100..105 没有进位出,Ci=0
rnz
output R
%endrep