lemans 0.0.0.pre → 0.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (56) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +9 -0
  3. data/LICENSE.txt +21 -0
  4. data/README.md +228 -0
  5. data/exe/lemans +17 -0
  6. data/lib/lemans/agents/base.rb +30 -0
  7. data/lib/lemans/agents/miniswen.rb +119 -0
  8. data/lib/lemans/agents/miniswen_installed.rb +67 -0
  9. data/lib/lemans/agents/nop.rb +15 -0
  10. data/lib/lemans/agents/oracle.rb +53 -0
  11. data/lib/lemans/agents.rb +21 -0
  12. data/lib/lemans/bench.rb +280 -0
  13. data/lib/lemans/cli/board_reporter.rb +135 -0
  14. data/lib/lemans/cli/progress_reporter.rb +67 -0
  15. data/lib/lemans/cli.rb +181 -0
  16. data/lib/lemans/clobber.rb +79 -0
  17. data/lib/lemans/environments/base.rb +55 -0
  18. data/lib/lemans/environments/daytona/retries.rb +49 -0
  19. data/lib/lemans/environments/daytona/sdk_tweaks.rb +50 -0
  20. data/lib/lemans/environments/daytona/shell.rb +142 -0
  21. data/lib/lemans/environments/daytona/snapshot_store.rb +163 -0
  22. data/lib/lemans/environments/daytona.rb +175 -0
  23. data/lib/lemans/environments.rb +16 -0
  24. data/lib/lemans/network_policy.rb +66 -0
  25. data/lib/lemans/patch.rb +70 -0
  26. data/lib/lemans/restore_paths.rb +21 -0
  27. data/lib/lemans/results/aggregate.rb +114 -0
  28. data/lib/lemans/results/cost_source.rb +13 -0
  29. data/lib/lemans/results/outcome.rb +36 -0
  30. data/lib/lemans/results/report.rb +149 -0
  31. data/lib/lemans/results/sorting.rb +24 -0
  32. data/lib/lemans/results/tally.rb +19 -0
  33. data/lib/lemans/results/usage.rb +24 -0
  34. data/lib/lemans/run.rb +152 -0
  35. data/lib/lemans/setup.rb +59 -0
  36. data/lib/lemans/setup_files.rb +36 -0
  37. data/lib/lemans/snapshot.rb +55 -0
  38. data/lib/lemans/task.rb +207 -0
  39. data/lib/lemans/tree_digest.rb +24 -0
  40. data/lib/lemans/trial.rb +187 -0
  41. data/lib/lemans/units.rb +44 -0
  42. data/lib/lemans/verifier/assets/eport-lemans.rb +36 -0
  43. data/lib/lemans/verifier/assets/lemans_minitest_reporter.rb +61 -0
  44. data/lib/lemans/verifier.rb +199 -0
  45. data/lib/lemans/version.rb +5 -0
  46. data/lib/lemans.rb +29 -0
  47. data/lib/miniswen/agent.rb +669 -0
  48. data/lib/miniswen/cli.rb +224 -0
  49. data/lib/miniswen/environment.rb +14 -0
  50. data/lib/miniswen/local.rb +42 -0
  51. data/lib/miniswen/ruby_llm.rb +42 -0
  52. data/lib/miniswen/testing.rb +134 -0
  53. data/lib/miniswen/trajectory.rb +110 -0
  54. data/lib/miniswen/version.rb +5 -0
  55. data/lib/miniswen.rb +48 -0
  56. metadata +160 -7
@@ -0,0 +1,19 @@
1
+ # frozen_string_literal: true
2
+
3
+ module Lemans
4
+ module Results
5
+ # One definition of the numbers everyone quotes — total, scored, invalid, solved — so the
6
+ # streaming summary and the report can never drift apart.
7
+ module Tally
8
+ def self.call(entries)
9
+ scored = entries.count { _1[:scored] }
10
+ {
11
+ total: entries.size,
12
+ scored: scored,
13
+ invalid: entries.size - scored,
14
+ solved: entries.count { _1[:reward].to_f >= 1.0 }
15
+ }
16
+ end
17
+ end
18
+ end
19
+ end
@@ -0,0 +1,24 @@
1
+ # frozen_string_literal: true
2
+
3
+ module Lemans
4
+ module Results
5
+ # What a trial spent. Cache tokens count separately, and the cost carries
6
+ # its own provenance — $0.00 is the figure most in need of auditing.
7
+ Usage = Data.define(:input_tokens, :output_tokens, :cached_tokens, :cost_usd, :steps, :cost_source) do
8
+ def self.zero
9
+ new(input_tokens: 0, output_tokens: 0, cached_tokens: 0, cost_usd: 0.0, steps: 0, cost_source: CostSource.none)
10
+ end
11
+
12
+ def to_h
13
+ {
14
+ input_tokens: input_tokens,
15
+ output_tokens: output_tokens,
16
+ cached_tokens: cached_tokens,
17
+ cost_usd: cost_usd,
18
+ steps: steps,
19
+ cost_source: cost_source&.to_h
20
+ }
21
+ end
22
+ end
23
+ end
24
+ end
data/lib/lemans/run.rb ADDED
@@ -0,0 +1,152 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "concurrent"
4
+ require "json"
5
+
6
+ module Lemans
7
+ # A whole run: every task, k attempts each, several trials in flight.
8
+ # Concurrency and resume exist because a sequential run is a day of wall clock.
9
+ class Run
10
+ Attempt = Data.define(:task, :model, :index)
11
+
12
+ # What ^C injects into workers instead of Interrupt, so the join loop can
13
+ # swallow its own stop signal while a user's second ^C stays distinguishable
14
+ class Shutdown < Exception; end # rubocop:disable Lint/InheritException
15
+
16
+ def initialize(bench:, tasks:, agent_name:, runs_dir:, attempts: 1, concurrency: 4,
17
+ resume: false, model: nil, backend: "daytona")
18
+ @bench = bench
19
+ @tasks = tasks
20
+ @agent_name = agent_name
21
+ @runs_dir = Pathname(runs_dir)
22
+ @attempts = attempts
23
+ @concurrency = concurrency
24
+ @resume = resume
25
+ @model = Array(model)
26
+ @backend = backend
27
+ end
28
+
29
+ # How many attempts this run will actually schedule — resume already
30
+ # subtracted the retired ones.
31
+ def total = pending.size
32
+
33
+ def call(&report)
34
+ begin
35
+ @runs_dir.mkpath
36
+ rescue SystemCallError => e
37
+ raise ConfigError, "cannot use runs directory #{@runs_dir}: #{e.message}"
38
+ end
39
+ queue = Queue.new
40
+ pending.shuffle.each { queue << _1 }
41
+ queue.close
42
+
43
+ results = Concurrent::Array.new
44
+ workers = Array.new(@concurrency) { Thread.new { drain(queue, results, &report) } }
45
+ workers.each(&:join)
46
+ raise @abort if @abort
47
+
48
+ summarize(results)
49
+ rescue Interrupt
50
+ queue&.clear
51
+ queue&.close
52
+ workers = Array(workers).select(&:alive?)
53
+ report&.call(:interrupted, { in_flight: workers.size })
54
+ workers.each { _1.raise(Shutdown) }
55
+
56
+ workers.each do |worker| # rubocop:disable Style/CombinableLoops
57
+ worker.join
58
+ rescue Shutdown
59
+ # ignore: our own stop signal coming back
60
+ end
61
+ summarize(results || []).merge(interrupted: true)
62
+ end
63
+
64
+ private
65
+
66
+ def pending
67
+ @pending ||= models.flat_map do |model|
68
+ @tasks.flat_map do |task|
69
+ completed = @resume ? completed_attempts(task, model) : 0
70
+ ((completed + 1)..@attempts).map { Attempt.new(task: task, model: model, index: _1) }
71
+ end
72
+ end
73
+ end
74
+
75
+ def models
76
+ return @model if @model.any?
77
+ return [nil] if @bench.agent.models.empty?
78
+
79
+ @bench.agent.models
80
+ end
81
+
82
+ # Counts only attempts that measured the same thing (agent, model, digests, scored) — otherwise
83
+ # editing bench.yml and resuming would let old-profile trials satisfy the new run.
84
+ def completed_attempts(task, model)
85
+ @runs_dir.glob("**/result.json").count { same_run?(_1, task, model) }
86
+ end
87
+
88
+ def same_run?(path, task, model)
89
+ result = JSON.parse(path.read, symbolize_names: true)
90
+
91
+ result[:task] == task.name &&
92
+ result[:agent] == @agent_name &&
93
+ result[:model] == (model || @bench.agent.model) &&
94
+ result[:profile_digest] == @bench.digest &&
95
+ result[:task_digest] == task.digest &&
96
+ result.dig(:outcome, :scored) == true
97
+ rescue JSON::ParserError, SystemCallError, IOError
98
+ false
99
+ end
100
+
101
+ # Any failure that escapes a trial poisons the run: stop scheduling, let
102
+ # in-flight trials finish, surface after the join.
103
+ def drain(queue, results, &)
104
+ # The TUI owns failure output; a dying worker must not spray stderr.
105
+ Thread.current.report_on_exception = false
106
+ while (attempt = queue.pop)
107
+ begin
108
+ results << run_attempt(attempt, &)
109
+ rescue StandardError => e
110
+ @abort ||= e
111
+ queue.clear
112
+ queue.close
113
+ end
114
+ end
115
+ end
116
+
117
+ def run_attempt(attempt)
118
+ trial = Trial.new(
119
+ task: attempt.task,
120
+ bench: @bench,
121
+ agent_name: @agent_name,
122
+ model: attempt.model,
123
+ backend: @backend,
124
+ runs_dir: @runs_dir
125
+ )
126
+
127
+ if block_given?
128
+ yield :started, { task: attempt.task.name, model: attempt.model || @bench.agent.model,
129
+ index: attempt.index, attempts: @attempts, trial: trial.id }
130
+ end
131
+
132
+ result = trial.run
133
+ if block_given?
134
+ yield :finished, {
135
+ task: attempt.task.name,
136
+ model: attempt.model || @bench.agent.model,
137
+ index: attempt.index,
138
+ outcome: result[:outcome][:name],
139
+ scored: result[:outcome][:scored],
140
+ reward: result[:reward],
141
+ duration_sec: result[:duration_sec],
142
+ detail: result[:outcome][:detail]
143
+ }
144
+ end
145
+ result
146
+ end
147
+
148
+ def summarize(results)
149
+ Results::Tally.call(results.map { { scored: _1.dig(:outcome, :scored) == true, reward: _1[:reward] } })
150
+ end
151
+ end
152
+ end
@@ -0,0 +1,59 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "shellwords"
4
+
5
+ module Lemans
6
+ # Makes one shared image into this task's sandbox at start-up: uploads the
7
+ # declared files and runs the setup commands. A failure here is never the model's.
8
+ class Setup
9
+ # Uploads land under one harness-owned directory, wiped once the steps have
10
+ # run: an agent whose first `ls /` finds the seed patch is reading the harness.
11
+ ROOT = "/lemans"
12
+ DIR = "#{ROOT}/setup".freeze
13
+
14
+ def initialize(commands:, task:, phase:, timeout_sec:)
15
+ @commands = commands
16
+ @task = task
17
+ @phase = phase
18
+ @timeout_sec = timeout_sec
19
+ end
20
+
21
+ def call(environment)
22
+ return if commands.empty? && files.empty?
23
+
24
+ files.each { |local, remote| environment.upload(local, remote) }
25
+ apply_seed(environment)
26
+ commands.each { environment.exec!(_1, timeout: timeout_sec) }
27
+ environment.exec!("rm -rf #{Shellwords.escape(ROOT)}")
28
+ end
29
+
30
+ private
31
+
32
+ attr_reader :commands, :task, :phase, :timeout_sec
33
+
34
+ # Folds a task's flat environment.patch into the workdir, then reseals the
35
+ # tree as a one-commit repo so `git log` does not point at the defect.
36
+ def apply_seed(environment)
37
+ return unless phase == :environment
38
+
39
+ seed = "#{DIR}/#{Task::FLAT_SEED}"
40
+ return unless files.any? { |_, remote| remote == seed }
41
+
42
+ workdir = Shellwords.escape(task.bench.environment.workdir)
43
+ environment.exec!(
44
+ "cd #{workdir} && git apply --binary --whitespace=nowarn #{Shellwords.escape(seed)} && " \
45
+ "rm -rf .git && git init -q && git add -A && " \
46
+ "git -c user.name=lemans -c user.email=lemans@localhost commit -qm 'Initial commit'",
47
+ timeout: timeout_sec
48
+ )
49
+ end
50
+
51
+ def files
52
+ @files ||= from(bench.root, bench.setup_files(phase)) + from(task.dir, task.setup_files(phase))
53
+ end
54
+
55
+ def from(root, paths) = paths.map { [root.join(_1), "#{DIR}/#{_1}"] }
56
+
57
+ def bench = task.bench
58
+ end
59
+ end
@@ -0,0 +1,36 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "pathname"
4
+
5
+ module Lemans
6
+ # The files a phase's setup steps consume. Paths are confined to the
7
+ # directory that declared them and must exist at load time.
8
+ module SetupFiles
9
+ PHASES = %i[environment verifier].freeze
10
+
11
+ def self.call(declared, root:, label:)
12
+ declared ||= {}
13
+ raise ConfigError, "#{label}: files must name a phase (#{PHASES.join(", ")})" unless declared.is_a?(Hash)
14
+
15
+ unknown = declared.keys.map(&:to_s) - PHASES.map(&:to_s)
16
+ raise ConfigError, "#{label}: files.#{unknown.first} is not a phase (#{PHASES.join(", ")})" if unknown.any?
17
+
18
+ root = Pathname(root)
19
+ PHASES.to_h do |phase|
20
+ [phase, Array(declared[phase.to_s]).map { checked(_1, phase, root: root, label: label) }.freeze]
21
+ end.freeze
22
+ end
23
+
24
+ def self.checked(path, phase, root:, label:)
25
+ raise ConfigError, "#{label}: files.#{phase} entry #{path.inspect} is not a path" unless path.is_a?(String) && !path.empty?
26
+ raise ConfigError, "#{label}: files.#{phase} entry #{path.inspect} must be relative to #{root}" if path.start_with?("/")
27
+
28
+ relative = Pathname(path).cleanpath
29
+ raise ConfigError, "#{label}: files.#{phase} entry #{path.inspect} must name a file inside #{root}" if relative.each_filename.include?("..") || relative.to_s == "."
30
+ raise ConfigError, "#{label}: files.#{phase} names #{path}, which is not a file" unless root.join(relative).file?
31
+
32
+ relative
33
+ end
34
+ private_class_method :checked
35
+ end
36
+ end
@@ -0,0 +1,55 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "shellwords"
4
+
5
+ module Lemans
6
+ # The sealed baseline of a task's graded surfaces: a git tree written before
7
+ # the agent's first turn, checked out over the live tree at verification.
8
+ class Snapshot
9
+ REMOTE_INDEX = "/tmp/lemans-baseline.idx"
10
+
11
+ TIMEOUT = 300
12
+
13
+ def initialize(environment, bench:, task:, timeout: TIMEOUT)
14
+ @environment = environment
15
+ @workdir = bench.environment.workdir
16
+ @paths = task.restore_paths
17
+ @timeout = timeout
18
+ @baseline = nil
19
+ end
20
+
21
+ # Sealing failure is an environment error: the agent has not run yet.
22
+ def capture!
23
+ return if paths.empty?
24
+
25
+ result = environment.exec(
26
+ "rm -f #{REMOTE_INDEX} && GIT_INDEX_FILE=#{REMOTE_INDEX} #{git} add -A && " \
27
+ "GIT_INDEX_FILE=#{REMOTE_INDEX} #{git} write-tree && rm -f #{REMOTE_INDEX}",
28
+ timeout: timeout
29
+ )
30
+ tree = result.output.to_s.lines.map(&:strip).reject(&:empty?).last
31
+ raise InfrastructureError, "could not seal the graded surfaces: #{result.output.to_s[0, 500]}" unless result.success? && /\A[0-9a-f]{40,64}\z/.match?(tree.to_s)
32
+
33
+ @baseline = tree
34
+ end
35
+
36
+ def restore! # rubocop:disable Naming/PredicateMethod
37
+ return true if paths.empty?
38
+ raise VerifierError, "restore is declared but no baseline was sealed" unless baseline
39
+
40
+ environment.exec(
41
+ "cd #{Shellwords.escape(workdir)} && #{git} cat-file -e #{baseline} && " \
42
+ "rm -rf -- #{escaped_paths} && #{git} checkout #{baseline} -- #{escaped_paths}",
43
+ timeout: timeout
44
+ ).success?
45
+ end
46
+
47
+ private
48
+
49
+ attr_reader :environment, :workdir, :paths, :timeout, :baseline
50
+
51
+ def git = "git -c safe.directory='*' -C #{Shellwords.escape(workdir)}"
52
+
53
+ def escaped_paths = Shellwords.join(paths)
54
+ end
55
+ end
@@ -0,0 +1,207 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "digest"
4
+ require "yaml"
5
+ require "pathname"
6
+
7
+ module Lemans
8
+ # One task: an instruction, an environment, a verifier, a solution.
9
+ # Anything lemans does not understand belongs under `metadata`, copied untouched.
10
+ class Task
11
+ INSTRUCTION = "instruction.md"
12
+ ENVIRONMENT_DIR = "environment"
13
+ TESTS_DIR = "tests"
14
+ SOLUTION_DIR = "solution"
15
+
16
+ FLAT_TEST = "verification_test.rb"
17
+ FLAT_SOLUTION = "solution.patch"
18
+ FLAT_SEED = "environment.patch"
19
+
20
+ FRONTMATTER = /\A---\n(.*?)\n---\n/m
21
+
22
+ # An image, either already published or built from a task's Dockerfile.
23
+ # Built images are named by content digest, so reuse is only ever of the identical thing.
24
+ class ImageSpec
25
+ attr_reader :reference, :dockerfile_path, :slug, :digest
26
+
27
+ def self.registry(reference) = new(reference: reference)
28
+
29
+ def self.dockerfile(path, slug:)
30
+ path = Pathname(path)
31
+ raise ConfigError, "no Dockerfile at #{path}" unless path.file?
32
+
33
+ new(dockerfile_path: path, slug: slug)
34
+ end
35
+
36
+ def initialize(reference: nil, dockerfile_path: nil, slug: nil)
37
+ @reference = reference
38
+ @dockerfile_path = dockerfile_path
39
+ @slug = slug
40
+ # Hashing the reference gives backends one answer to "is this the same image" either way.
41
+ @digest = built? ? TreeDigest.call(context_dir) : Digest::SHA256.hexdigest(reference.to_s)
42
+ freeze
43
+ end
44
+
45
+ def built? = !dockerfile_path.nil?
46
+
47
+ def context_dir = dockerfile_path&.dirname
48
+
49
+ def name
50
+ built? ? "lemans-#{digest[0, 32]}" : reference
51
+ end
52
+
53
+ def to_s = name
54
+ end
55
+
56
+ attr_reader :dir, :name, :description, :difficulty, :tags, :metadata, :bench, :digest
57
+
58
+ def self.load(dir, bench:)
59
+ dir = Pathname(dir)
60
+ new(frontmatter(dir), dir: dir, bench: bench)
61
+ end
62
+
63
+ def self.frontmatter(dir)
64
+ content = dir.join(INSTRUCTION).read
65
+ match = content.match(FRONTMATTER)
66
+ unless match
67
+ # Opens like frontmatter but never matches: silently dropping every
68
+ # declared key (and leaking the raw block to the agent) is worse than
69
+ # refusing. CRLF endings and a missing final newline are the usual causes.
70
+ raise ConfigError, "#{dir.join(INSTRUCTION)}: frontmatter opens with --- but never closes" if
71
+ content.start_with?("---")
72
+
73
+ return {}
74
+ end
75
+ config = YAML.safe_load(match[1], aliases: true) || {}
76
+ raise ConfigError, "#{dir.join(INSTRUCTION)}: frontmatter must be a mapping" unless config.is_a?(Hash)
77
+
78
+ config
79
+ rescue Errno::ENOENT
80
+ {} # fallback to validate!
81
+ rescue Psych::Exception => e
82
+ raise ConfigError, "#{dir.join(INSTRUCTION)}: #{e.message}"
83
+ end
84
+
85
+ def initialize(config, dir:, bench:)
86
+ @dir = Pathname(dir)
87
+ @bench = bench
88
+ @name = config["name"] || @dir.basename.to_s
89
+ @description = config["description"]
90
+ @difficulty = config["difficulty"]
91
+ @tags = Array(config["tags"]).freeze
92
+ @metadata = (config["metadata"] || {}).freeze
93
+ @files = SetupFiles.call(config["files"], root: @dir, label: @dir)
94
+ @restore = config.key?("restore") ? RestorePaths.call(config["restore"], label: "#{@dir}: restore") : nil
95
+
96
+ validate!(config)
97
+ # Recorded on every result: without it a reward cannot say which task version it measured.
98
+ @digest = TreeDigest.call(@dir)[0, 16]
99
+ freeze
100
+ end
101
+
102
+ # The story alone: frontmatter is for the harness, never for the agent.
103
+ def instruction = instruction_path.read.sub(FRONTMATTER, "")
104
+
105
+ def instruction_path = dir.join(INSTRUCTION)
106
+
107
+ def environment_context = dir.join(ENVIRONMENT_DIR)
108
+
109
+ def environment_dockerfile = environment_context.join("Dockerfile")
110
+
111
+ # Stays on the harness side while the agent works; uploaded into the sandbox only at verification.
112
+ def tests_dir = dir.join(TESTS_DIR)
113
+
114
+ # [absolute, remote-relative] pairs.
115
+ def test_files
116
+ if tests_dir.directory?
117
+ expand(tests_dir)
118
+ else
119
+ flat(FLAT_TEST)
120
+ end
121
+ end
122
+
123
+ def solution_files
124
+ if solution_context.directory?
125
+ expand(solution_context)
126
+ else
127
+ flat(FLAT_SOLUTION)
128
+ end
129
+ end
130
+
131
+ def environment_image
132
+ if bench.environment.image
133
+ ImageSpec.registry(bench.environment.image)
134
+ else
135
+ ImageSpec.dockerfile(environment_dockerfile, slug: name)
136
+ end
137
+ end
138
+
139
+ def setup_files(phase)
140
+ declared = @files.fetch(phase.to_sym, [])
141
+ seed = Pathname(FLAT_SEED)
142
+ return declared unless phase.to_sym == :environment && dir.join(seed).file? && !declared.include?(seed)
143
+
144
+ declared + [seed]
145
+ end
146
+
147
+ def solution_context = dir.join(SOLUTION_DIR)
148
+
149
+ def solution? = solution_files.any?
150
+
151
+ def restore_paths = @restore || bench.verifier.restore_paths
152
+
153
+ def to_h
154
+ {
155
+ name: name,
156
+ description: description,
157
+ difficulty: difficulty,
158
+ tags: tags,
159
+ metadata: metadata
160
+ }.compact
161
+ end
162
+
163
+ private
164
+
165
+ # Dotfiles included, matching TreeDigest: the files a digest records are
166
+ # exactly the files that ship.
167
+ def expand(root)
168
+ root.glob("**/*", File::FNM_DOTMATCH).select(&:file?).map { [_1, _1.relative_path_from(root).to_s] }
169
+ end
170
+
171
+ def flat(filename)
172
+ path = dir.join(filename)
173
+ path.file? ? [[path, filename]] : []
174
+ end
175
+
176
+ def validate!(config)
177
+ raise ConfigError, "#{dir}: #{INSTRUCTION} is required" unless instruction_path.file?
178
+
179
+ refuse_bench_collisions!
180
+
181
+ raise ConfigError, "#{dir}: #{ENVIRONMENT_DIR}/Dockerfile is required when bench.yml names no shared image" unless bench.environment.image || environment_dockerfile.file?
182
+
183
+ if test_files.empty?
184
+ raise ConfigError, "#{dir}: #{TESTS_DIR}/ or a flat #{FLAT_TEST} is required — " \
185
+ "the verifier uploads it at verification time"
186
+ end
187
+
188
+ return unless config.key?("overrides")
189
+
190
+ declared = config["overrides"]
191
+ named = declared.is_a?(Hash) && declared.any? ? " (#{declared.keys.join(", ")})" : ""
192
+ raise ConfigError, "#{dir}: a task cannot override the frozen profile#{named} — " \
193
+ "what has to vary belongs in bench.yml, where it varies for every trial"
194
+ end
195
+
196
+ # Collisions would be resolved by upload order, so a task never gets to shadow the bench-wide copy.
197
+ def refuse_bench_collisions!
198
+ SetupFiles::PHASES.each do |phase|
199
+ shadowed = @files.fetch(phase) & bench.setup_files(phase)
200
+ next if shadowed.empty?
201
+
202
+ raise ConfigError, "#{dir}: files.#{phase} names #{shadowed.first}, which #{bench.path.basename} " \
203
+ "already ships to every task"
204
+ end
205
+ end
206
+ end
207
+ end
@@ -0,0 +1,24 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "digest"
4
+
5
+ module Lemans
6
+ # A directory's contents reduced to one number. Paths hash alongside
7
+ # contents, and dotfiles are included — glob skips them by default.
8
+ module TreeDigest
9
+ def self.call(dir)
10
+ dir = Pathname(dir)
11
+ sha = Digest::SHA256.new
12
+ dir.glob("**/*", File::FNM_DOTMATCH).sort.each do |entry|
13
+ next unless entry.file?
14
+
15
+ path = entry.relative_path_from(dir).to_s
16
+ contents = entry.binread
17
+ sha << [path.bytesize, contents.bytesize].pack("Q>Q>")
18
+ sha << path
19
+ sha << contents
20
+ end
21
+ sha.hexdigest
22
+ end
23
+ end
24
+ end