Test.hs

Centralized test suite validating configuration files, metadata databases, and module unit tests

Pathbuild/Test.hs

LanguageHaskell

Lines299

Sourcebuild/Test.hs
at 406d3e423

Read this when Use this page when tracing the Haskell build pipeline, generators, metadata code, or backend utility behavior around Test.

Overview

Test.hs is the unified test harness for the gwern.net Haskell backend. Rather than scattering tests across individual modules (which would incur per-call overhead and risk quadratic behavior when validating lists), this module consolidates all validation into a single testAll entry point that runs during the build process.

The test suite validates three categories of concerns: (1) configuration file integrity—checking for duplicates, valid URLs, proper formats, and absence of cycles in rewrite databases; (2) regex pattern validity—ensuring all regex strings compile without errors; and (3) module-level unit tests—verifying that individual functions like cleanAbstractsHTML, titleCase, and tooltipToMetadata produce expected outputs. The design philosophy prioritizes catching configuration errors early, since many configs are hand-edited and prone to typos, duplicates, or format mistakes.

Test.hs imports from virtually every other module in the codebase, making it both a validation layer and an implicit documentation of cross-module dependencies. When tests fail, they print red error messages identifying exactly which config or test case failed; when they pass, they print green status messages with counts of verified items.

---

Public API

testAll :: IO ()

Main entry point that runs all tests sequentially. Prints colored output indicating pass/fail status.

Called by: hakyll.hs (build system), manual invocation Calls: Every test function in the module, plus functions from 20+ other modules

testConfigs :: Int

Pure function that forces centralized configuration sanity checks and returns an accounting value used by testAll output. The source comments warn not to interpret this Int as a pass/fail error count; failed checks abort or report through their own validation paths.

Called by: testAll Calls: isUniqueList, isUniqueKeys, isUniqueAll, isCycleLess, ensure, isDomainT, isURL* family

testXOTD :: IO Int

Validates the "X of the Day" quote and site databases for format correctness and uniqueness.

Called by: testAll Calls: XOTD.readTTDB, isUniqueKeys3, ensure

testRegexPatterns :: [String] -> IO ()

Attempts to compile each regex pattern, printing errors for any that fail.

Called by: testAll Calls: makeRegexM

---

Internal Architecture

Test Organization

testAll
├── Link Icon Tests (linkIconTest)
├── Interwiki Tests (interwikiTestSuite, interwikiCycleTestSuite)
├── Config Validation (testConfigs)
│   ├── Uniqueness checks (isUniqueList, isUniqueKeys, isUniqueAll)
│   ├── Format validation (ensure with predicates)
│   └── Cycle detection (isCycleLess)
├── Regex Validation (testRegexPatterns)
├── Metadata Database Tests
│   ├── readLinkMetadata
│   ├── readArchiveMetadataAndCheck
│   └── URL hash uniqueness
├── Unit Test Suites
│   ├── cleanAbstractsHTML
│   ├── titleCase
│   ├── authorCollapse
│   ├── tooltipToMetadata
│   └── many more...
└── Live Tests (Wikipedia API calls)

Validation Helpers from Unique Module

• isUniqueList: Checks a list has no duplicate elements
• isUniqueKeys: Checks a list of tuples has unique first elements
• isUniqueAll: Checks both keys and values are unique
• isUniqueKeys3/isUniqueKeys4: For 3-tuples and 4-tuples

The ensure Pattern

ensure :: (Show a, NFData a) => String -> String -> (a -> Bool) -> [a] -> [a]

Checks every item of a list against a predicate (defined in Utils.hs). Used extensively:

ensure "Test.linkIDOverrides" "isURLAny" isURLAny Config.LinkID.linkIDOverrides

Returns the list unchanged if all items pass; if any item fails, evaluation aborts with an error naming the location, the property, and the offending item.

---

Key Patterns

Centralized Config Testing

The design deliberately avoids testing configs at point-of-use:

-- Run pure config sanity checks in one place rather than in the hot paths
-- that consume each config. Many checks scan whole lists or maps. Embedding
-- them in link conversion, filtering, or cleanup functions would repeat that
-- scan for every input element, quietly turning an otherwise-linear pass
-- into an accidental quadratic one.

This means adding a new config requires manually adding its test to testConfigs.

Colored Output Protocol

printGreen ("Testing link icon matches…" :: String)
unless (null linkIconTest) $ printRed ("Link icon rules have errors in: " ++ show linkIconTest)

Green messages indicate test categories being run; red indicates failures. Success is silent beyond the category announcement.

Cycle Detection in Rewrite Databases

Several configs (author canonicalization, tag rewrites, interwiki redirects) could create infinite loops if entries reference each other cyclically:

length $ isCycleLess Config.Metadata.Author.canonicals
length $ isCycleLess Config.Tags.tagsShort2LongRewrites

Live API Tests

The Wikipedia tests actually hit the Wikipedia API:

a <- Interwiki.isWPArticle False "https://en.wikipedia.org/wiki/George_Washington_XYZ"
b <- Interwiki.isWPArticle False "https://en.wikipedia.org/wiki/George_Washington"
c <- Interwiki.isWPDisambig "Mercury"

These verify that article-existence and disambiguation checking works correctly.

---

Configuration

Test.hs itself has no configuration. It tests configs from these modules:

Config Module	What's Tested
Config.GenerateSimilar	URL blacklists are valid URLs
Config.Interwiki	Rewrites, redirects, quote overrides
Config.LinkArchive	Whitelist matches, localization test cases
Config.LinkIcon	Icon rules, blacklists
Config.LinkLive	Domain whitelists/blacklists
Config.LinkSuggester	Bad anchor strings, whitelists
Config.Tags	Tag mappings, rewrites
Config.Typography	Title case rules, date ranges
Config.XOfTheDay	Site/quote databases
Config.Inflation	Bitcoin exchange rates, test cases
Config.LinkID	ID overrides, affiliation anchors
Config.LinkMetadata	URI validation exemptions and previewable extension lists
Config.Utext	Raw Markdown-to-Utext rendering cases
Config.Metadata.*	Author, Format, Title configs
Config.Misc	Tooltips, cycles, arxiv abstracts
Config.Paragraph	Paragraph whitelist

---

Integration Points

Imports (Tests From)

Test.hs imports test functions from these modules:

• Annotation.tooltipToMetadata, testGuessAuthorDate
• Cycle.testCycleDetection
• Inflation.inflationDollarTestSuite
• Interwiki.interwikiTestSuite, interwikiCycleTestSuite
• LinkArchive.testLinkRewrites
• GenerateSimilar.generateSimilarTestSuite
• LinkIcon.linkIconTest
• LinkLive.linkLiveTest, linkLivePrioritize
• Tags.testTags
• Typography.titleCaseTest
• Config.LinkMetadata.fileTranscludesTest (run against LinkMetadata.generateFileTransclusionBlock)
• Utext.rawMarkdown2Utext, rawMarkdown2UtextStyled
• Metadata.Author.authorCollapseTest, cleanAuthorsTest
• Metadata.Format.printDoubleTestSuite, cleanAbstractsHTMLTest
• Metadata.Date.dateRangeDurationTestCasesTestsuite

Working Directory

testAll = do Config.Misc.cd  -- Changes to project root

Tests assume they run from the project root directory.

Exit Behavior

Unit-test failures don't exit—they print red errors and continue, so the build system must check output separately. However, the pure config sanity checks (ensure, the Unique helpers, isCycleLess) abort with error when forced, so a bad config entry does crash the test run.

---

What Is Tested

Configuration Integrity

1. Uniqueness: No duplicate entries in lists, no duplicate keys in maps
2. URL Validity: URLs are actually URLs (isURL, isURLT, isURLAny)
3. Domain Validity: Domain strings are valid domains (isDomainT)
4. Cycle Freedom: Rewrite chains don't loop infinitely
5. Format Correctness: HTML identifiers start with letters, dates are valid dates

Regex Validity

All regex patterns from configs are compiled to catch syntax errors:

• footnoteRegex, sectionAnonymousRegex, badUrlRegex
• Tag rewrite regexes
• Author cleanup regexes
• HTML rewrite regexes
• Arxiv abstract regexes

Unit Test Suites

Function	Config Source
balanced	balancedBracketTestCases
cleanAbstractsHTML	htmlRewriteTestCases
extractTwitterUsername	extractTwitterUsernameTestSuite
tooltipToMetadata	tooltipToMetadataTestcases
titleCase	titleCaseTestCases
authorCollapse	authorCollapseTestCases
fileTranscludes	Generated from metadata

Database Integrity

• Link Metadata: URL hashes are unique, IDs are valid
• Archive Metadata: Checked via readArchiveMetadataAndCheck
• X-of-the-Day: Quote/site DBs have valid format

---

See Also

• hakyll.hs - Build system that invokes tests during SLOW builds
• Utils.hs - Utility functions with unit tests
• sync.sh - Build orchestrator controlling test execution
• LinkMetadata.hs - Metadata database being validated
• Typography.hs - Typography functions with unit tests
• Cycle.hs - Cycle detection tested here