Draft PR to store changes to my blog post by robbie-c · Pull Request #17877 · PostHog/posthog.com

robbie-c · 2026-06-24T00:26:14Z

Changes

See the original https://docs.google.com/document/d/1d0J9OUwxN7uCD9q7TMU8P5S91IOg3Aoe9TTP_OKVCFI/edit?tab=t.0
See #17681

Checklist

I've read the docs and/or content style guides.
Words are spelled using American English
Use relative URLs for internal links
I've checked the pages added or changed in the Vercel preview build
If I moved a page, I added a redirect in vercel.json

My version on top of #17681: keeps the original intro and voice, adopts the section headings, links, and restructured lists, adds new content (logical-vs-physical-layout rationale, ShrinkRay, coverage-guided generation), and a few copy-edit passes.

github-actions · 2026-06-24T00:26:48Z

Deploy preview

Status	Details	Updated (UTC)
🟢 Ready	View preview	Jun 24, 2026 12:37AM

github-actions · 2026-06-24T00:27:11Z

Vale prose linter → found 0 errors, 19 warnings, 0 suggestions in your markdown

Full report → Copy the linter results into an LLM to batch-fix issues.

Linter being weird? Update the rules!

contents/blog/sql-parser.md — 0 errors, 19 warnings, 0 suggestions

Line	Severity	Message	Rule
16:51	warning	'clacky' is a possible misspelling.	`PostHogBase.Spelling`
22:73	warning	'autoresearch' is a possible misspelling.	`PostHogBase.Spelling`
28:70	warning	'transpile' is a possible misspelling.	`PostHogBase.Spelling`
33:37	warning	Capitalize 'Product Analytics' for PostHog's product. Use 'product analytics' for the general industry concept.	`PostHogBase.ProductNames`
33:56	warning	Capitalize 'Session Replay' for PostHog's product. Use 'session replay' for the general industry concept.	`PostHogBase.ProductNames`
33:72	warning	Capitalize 'Error Tracking' for PostHog's product. Use 'error tracking' for the general industry concept.	`PostHogBase.ProductNames`
33:151	warning	'transpilation' is a possible misspelling.	`PostHogBase.Spelling`
33:285	warning	'untrusted' is a possible misspelling.	`PostHogBase.Spelling`
35:27	warning	'transpilation' is a possible misspelling.	`PostHogBase.Spelling`
35:132	warning	'transpiled' is a possible misspelling.	`PostHogBase.Spelling`
37:4	warning	'Generating our parser with ANTLR' heading should be in sentence case, and product names should be capitalized.	`PostHogBase.SentenceCase`
41:134	warning	'declaratively' is a possible misspelling.	`PostHogBase.Spelling`
45:48	warning	'lookahead' is a possible misspelling.	`PostHogBase.Spelling`
53:162	warning	'lookahead' is a possible misspelling.	`PostHogBase.Spelling`
67:40	warning	'transpiler' is a possible misspelling.	`PostHogBase.Spelling`
71:93	warning	'codegen' is a possible misspelling.	`PostHogBase.Spelling`
75:182	warning	'lookahead' is a possible misspelling.	`PostHogBase.Spelling`
75:237	warning	'lookahead' is a possible misspelling.	`PostHogBase.Spelling`
87:13	warning	'anonymized' is a possible misspelling.	`PostHogBase.Spelling`

robbie-c · 2026-06-24T00:28:58Z

+
+After the success of using agents to [improve query performance through autoresearch](/blog/karpathy-autoresearch-query-engine-bug), I wanted to try something more ambitious.
+
+I ran multiple long-running Claude Code sessions in parallel, and the result was 16K lines of "hand"-rolled parser code, 5K lines of tooling, and a few more K of tests.


I kept this instead of the stat about number of queries. My rationale is that this actually isn't a story about doing something at scale, it's more about doing something with AI coding that's hard with some interesting computer science

robbie-c · 2026-06-24T00:29:35Z

+
+Hypothesis will "reduce" test cases for you, turning them into a minimal reproduction, but I couldn’t use that with SQL from other sources. For those I used [ShrinkRay](https://github.com/DRMacIver/shrinkray) instead.
+
+Later on, I added code-coverage-guided test case generation, which gives a better distribution of generated SQL. With coverage feedback, the generator can tell which constructs it hasn't exercised yet and bias towards those. This wasn't necessary to hit 100% accuracy on a production corpus, but it did help me find some very subtle test cases.


this is new, I did this since I wrote the original

robbie-c · 2026-06-24T00:29:41Z

+
+The two parallel parser approaches shared their regression suites, so any failing test case found in one session was shared with the other.
+
+Hypothesis will "reduce" test cases for you, turning them into a minimal reproduction, but I couldn’t use that with SQL from other sources. For those I used [ShrinkRay](https://github.com/DRMacIver/shrinkray) instead.


this is new, I did this since I wrote the original

github-actions · 2026-06-24T00:36:56Z

Bundle report

Total JS (gzip)

6.21 MiB (no change)

Eager graph (static-import closure per entrypoint)

Entrypoint	Eager size	Budget	Modules
✅ `app`	24.13 MiB (no change)	report-only	5505

Largest modules in the app closure

Module	Size
`css ./node_modules/.pnpm/css-loader@5.2.7_webpack@5.101.3/node_modules/css-loader/dist/cjs.js??ruleSet[1].rules[8].oneOf[1].use[1]!./node_modules/.pnpm/postcss-loader@4.3.0_postcss@8.5.6_webpack@5.101.3/node_modules/postcss-loader/dist/cjs.js??ruleSet[1].rules[8].oneOf[1].use[2]!./src/styles/global.css`	710.3 KiB
`./src/components/Stickers/Stickers.tsx`	696.4 KiB
`./.cache/caches/gatsby-plugin-mdx/mdx-scopes-dir/31a094f140f119e73085d847ae81b99b.js + 2 modules`	531.3 KiB
`./node_modules/.pnpm/@radix-ui+react-icons@1.3.2_react@18.3.1/node_modules/@radix-ui/react-icons/dist/react-icons.esm.js`	481.4 KiB
`./node_modules/.pnpm/@codemirror+view@6.38.2/node_modules/@codemirror/view/dist/index.js`	458.1 KiB
`./node_modules/.pnpm/rehype-raw@7.0.0/node_modules/rehype-raw/lib/index.js + 29 modules`	395.1 KiB
`./node_modules/.pnpm/@posthog+icons@0.36.6_react-dom@18.3.1_react@18.3.1__react@18.3.1/node_modules/@posthog/icons/dist/posthog-icons.cjs.js`	364.8 KiB
`./node_modules/.pnpm/@posthog+icons@0.36.6_react-dom@18.3.1_react@18.3.1__react@18.3.1/node_modules/@posthog/icons/dist/posthog-icons.es.js`	354.8 KiB
`./src/hooks/useCustomers.tsx + 54 modules`	353.9 KiB
`./node_modules/.pnpm/react-markdown@8.0.7_@types+react@16.14.66_react@18.3.1/node_modules/react-markdown/lib/react-markdown.js + 88 modules`	351.4 KiB
`./node_modules/.pnpm/cloudinary-core@2.14.0_lodash@4.17.21/node_modules/cloudinary-core/cloudinary-core.js`	281.9 KiB
`./node_modules/.pnpm/@codesandbox+sandpack-react@2.20.0_react-dom@18.3.1_react@18.3.1__react@18.3.1/node_modules/@codesandbox/sandpack-react/dist/index.mjs`	266.6 KiB
`./src/components/ProductComparisonTable/index.tsx + 114 modules`	264.0 KiB
`./node_modules/.pnpm/d3@7.9.0/node_modules/d3/src/index.js + 208 modules`	247.4 KiB
`./src/components/Pricing/PricingSlider/Slider.tsx + 87 modules`	239.9 KiB

Eager-graph budgets are report-only until a baseline is established. Sizes are gzip of public/**/*.js; eager size is webpack module source bytes.

ivanagas and others added 2 commits June 16, 2026 16:26

sql parser

afe0111

Blog: edits to the SQL parser post

4525d27

My version on top of #17681: keeps the original intro and voice, adopts the section headings, links, and restructured lists, adds new content (logical-vs-physical-layout rationale, ShrinkRay, coverage-guided generation), and a few copy-edit passes.

robbie-c commented Jun 24, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Draft PR to store changes to my blog post#17877

Draft PR to store changes to my blog post#17877
robbie-c wants to merge 2 commits into
masterfrom
sql-parser-robbie-edits

robbie-c commented Jun 24, 2026

Uh oh!

github-actions Bot commented Jun 24, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jun 24, 2026

Uh oh!

robbie-c Jun 24, 2026

Uh oh!

robbie-c Jun 24, 2026

Uh oh!

robbie-c Jun 24, 2026

Uh oh!

github-actions Bot commented Jun 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants


		After the success of using agents to [improve query performance through autoresearch](/blog/karpathy-autoresearch-query-engine-bug), I wanted to try something more ambitious.

		I ran multiple long-running Claude Code sessions in parallel, and the result was 16K lines of "hand"-rolled parser code, 5K lines of tooling, and a few more K of tests.


		Hypothesis will "reduce" test cases for you, turning them into a minimal reproduction, but I couldn’t use that with SQL from other sources. For those I used [ShrinkRay](https://github.com/DRMacIver/shrinkray) instead.

		Later on, I added code-coverage-guided test case generation, which gives a better distribution of generated SQL. With coverage feedback, the generator can tell which constructs it hasn't exercised yet and bias towards those. This wasn't necessary to hit 100% accuracy on a production corpus, but it did help me find some very subtle test cases.


		The two parallel parser approaches shared their regression suites, so any failing test case found in one session was shared with the other.

		Hypothesis will "reduce" test cases for you, turning them into a minimal reproduction, but I couldn’t use that with SQL from other sources. For those I used [ShrinkRay](https://github.com/DRMacIver/shrinkray) instead.

Conversation

robbie-c commented Jun 24, 2026

Changes

Checklist

Uh oh!

github-actions Bot commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Deploy preview

Uh oh!

github-actions Bot commented Jun 24, 2026

Uh oh!

robbie-c Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

robbie-c Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

robbie-c Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Jun 24, 2026

Bundle report

Total JS (gzip)

Eager graph (static-import closure per entrypoint)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented Jun 24, 2026 •

edited

Loading