Reset repository as the Go foundation for the software factory - #2
Open
ihtkas wants to merge 4 commits into
Open
Reset repository as the Go foundation for the software factory#2ihtkas wants to merge 4 commits into
ihtkas wants to merge 4 commits into
Conversation
This repository previously held algorithm and data-structure exercises. That work shares no lineage with what go-libs becomes, so it is removed rather than adapted. It is preserved on the archive/algorithms branch. go-libs is now a component library: well-designed, independently usable Go packages that applications import as ordinary dependencies. The premise is that SaaS forces multi-tenancy, distribution, and hyperscale onto every customer, who then pays for generality they never needed. With the cost of bespoke software collapsing, a company can instead own production-grade software shaped to how it actually works and self-host it. These packages are the foundation that makes that practical. - Remove graph, sort, string, and tree exercise packages - Remove the placeholder GitHub Actions workflow, which only echoed on issue comments - Add CLAUDE.md: thesis, architecture, engineering standards, design process, non-goals, and a record of deliberately deferred decisions - Rewrite README.md as a public statement of intent - Raise the go directive from 1.15 to 1.26 No packages are added here. Each one is designed before it is built, and only when it is clearly better than what a caller would otherwise write or import. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CLAUDE.md and README.md argued their case at length where stating it once was enough. Documents here should convey meaning briefly, not build a case. Removed the restated thesis in the architecture and non-goals sections, the expanded justifications under each engineering standard, and the ASCII dependency diagram that the adjacent bullets already described. Every rule, constraint, and deferred decision is retained. Adds one standard: prose in this repo states the decision and stops. CLAUDE.md 10.3K to 5.6K, README.md 1.3K to 0.9K. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
"It generates no code and boots no application" was too absolute. It read as forbidding any package that generates code, which would rule out query codegen, client generation, and mock generation — all legitimate library work that does not make this a framework. The invariant is control, not capability: applications call into go-libs, never the reverse, and it never owns main. A generator the developer runs deliberately does not breach that. Also records that generation is preferred over reflection. Explicit codegen emitting readable, checked-in Go is how Go avoids runtime magic, so it expresses the no-magic rule rather than violating it. Generators live under cmd/. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds .claude/skills/go-review: a pre-merge review gate for Go code, backed by a pinned verbatim snapshot of the Uber Go Style Guide (uber-go/guide 37c45b52bf9a, retrieved 2026-07-30, Apache 2.0). NOTICE carries the attribution; the snapshot header records how to refresh it. Authority is CLAUDE.md first, the Uber guide second. Two rules where this repo deliberately departs from Uber are recorded as overrides so reviews do not relitigate them: - go.uber.org/atomic — rejected. It breaches our dependency policy, and sync/atomic has had typed atomics since Go 1.19. - Start enums at one — rejected as a default. It collides with useful zero values; it applies only where an unset value is genuinely invalid. Standards updated where reading the guide showed real gaps: - Errors: wrapping with %w was stated unconditionally. It makes the wrapped error part of the contract, so it is now conditional on whether callers should match it. - API: copy slices and maps at boundaries; no embedding in exported structs. - State: new — no mutable package-level state, and constraints on init(). A global in a library makes every consumer's tests order-dependent with no way to opt out. - Testing: failure messages identify the case and report got against want. - Style: initialism casing, no Get prefix, package and receiver naming. Effective Go and the Google style guide were evaluated and dropped; the Uber guide is the sole external reference. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this does
Clears out the algorithm and data-structure exercises this repository was holding, and re-establishes
go-libsas the Go foundation for the software factory.The previous contents share no lineage with where this is going, so they are removed rather than adapted. Nothing is lost — the full prior history is preserved on the
archive/algorithmsbranch.Why
A SaaS vendor writes one system for every customer at once. That forces multi-tenancy, distribution, and hyperscale into the design, and every customer pays for that generality — in fees, in integration work, and in workflows bent to fit someone else's model. Most companies never needed the complexity they are paying for.
The cost of building bespoke software has collapsed. A company can now own production-grade software shaped to how it actually works, rivalling the SaaS product it would otherwise rent, and self-host it. What is missing is not the ability to generate code — it is a foundation worth generating code against.
go-libsis that foundation: independently usable Go packages that applications import as ordinary dependencies. Not a framework.Changes
graph/,sort/,string/,tree/— 25 exercise files.github/workflows/build.yml— a placeholder that only echoed on issue commentsCLAUDE.md— thesis, architecture, engineering standards, design process, non-goals, deferred decisionsREADME.md— public statement of intentgodirective1.15→1.26Architecture established
One
go.modfor the whole tree, revisited only if the tree becomes hard to manage. A strict dependency spine:coredepends only on the standard library — nothing else, evercoreand stdlibThe lateral-import rule is load-bearing: it keeps this a parts bin rather than a framework, and keeps a future multi-module split mechanical rather than archaeological.
No packages are added here
By design. Each package begins with a written design, reviewed before implementation, and none is built autonomously. Breadth is not a goal — an adequate library in this position is worse than none, because it gets embedded in real systems before anyone notices it was not thought through. When the honest answer is "use the standard library," no package gets written.
CLAUDE.mdalso records what was deliberately deferred — operational shape, multi-module split, and which packages to build — along with the trigger for revisiting each, so those stay parked on purpose rather than getting quietly re-litigated.🤖 Generated with Claude Code