Conversation
Complete Go modules reference covering module paths, versions, go.mod directives, go commands, proxies, version control, and authentication.
…slations Fix 18 files where Vietnamese prose used inconsistent terminology: blog/: - 16years.md: "garbage collector" → "bộ gom rác" (2x) - contributor-workshop.md: "contributor" → "người đóng góp" (5x), "repository" → "kho lưu trữ" - contributors-summit-2019.md: "contributor" → "người đóng góp" (4x), "open-source" → "mã nguồn mở" - contributors-summit.md: "contributor" → "người đóng góp" (multiple), "tooling" → "hệ thống công cụ" - cover.md: "tooling" → "hệ thống công cụ" (2x) - generic-slice-functions.md: "garbage collector" → "bộ gom rác" (3x) - gofix.md: "false positive" → "dương tính giả" - open-source.md: "garbage collector" → "bộ gom rác", "contributor" → "người đóng góp" (14x) - pgo-preview.md: "repository" → "kho lưu trữ" (2x), PGO term fix (4x) - pgo.md: "garbage collector" → "bộ gom rác", "repository" → "kho lưu trữ" (2x), PGO term fix (5x) - qihoo.md: "garbage collection" → "bộ gom rác" - survey2023-q1-results.md: "repository" → "kho lưu trữ" - wire.md: "tooling" → "hệ thống công cụ" doc/: - go1.12.md: "garbage collector" → "bộ gom rác" - go1.22.md: "Profile-Guided Optimization" → "tối ưu hóa dựa trên hồ sơ thực thi" - go1.23.md: "Profile Guided Optimization" → "tối ưu hóa dựa trên hồ sơ thực thi" - go1.26.md: "garbage collector/Garbage collector" → "bộ gom rác" (6x) solutions/: - mercadolibre.md: "repository" → "kho lưu trữ" (3x)
- contributor: keep unchanged (do not translate to "người đóng góp") - repository: keep unchanged (do not translate to "kho lưu trữ") Reverted in 8 files: blog/contributor-workshop.md, contributors-summit-2019.md, contributors-summit.md, open-source.md, pgo.md, pgo-preview.md, survey2023-q1-results.md, solutions/mercadolibre.md Updated GLOSSARY.md to mark both terms as "keep unchanged".
Sync translations with golang/website upstream commits: - e2752c0 (_content: add more documentation for 'go tool' features in ref/mod) - c961c3d (_content: in managing dependencies doc '-u' flag and mention '-modfile') Changes in _content_vi/ref/mod.md: - tool directive: add "Kể từ Go 1.24," prefix - go get usage: add -tool flag - go get flags: add -tool flag description - go get -d: fix tense (sẽ luôn → luôn) - go install constraints: conditionalize on version-suffix args - go install file paths bullet: note go install tool works without version suffix - go install last para: add go install tool for module tools - Add new `go tool` section with usage, examples, and description Changes in _content_vi/doc/modules/managing-dependencies.md: - Remove unnecessary -u from `go get -u tool` → `go get tool` - Add equivalence note: go get tool@upgrade
Quote summary field to fix parse error: unquoted colon-space inside a YAML scalar value causes the parser to interpret it as a mapping.
Full translation of /ref/spec (8928 lines) into _content_vi/ref/spec.html. All prose translated to Vietnamese; code blocks, identifiers, and HTML attributes preserved unchanged.
Translate Go language specification to Vietnamese
- content.go: embed _content_vi alongside _content; Content() now returns an overlayFS (vi → en) so Vietnamese translations take priority while missing assets (css/js/fonts/images) fall back to English originals. - server.go: auto-detect _content_vi before _content for live-reload in local dev; NewHandler constructs the same overlay when contentDir ends with _content_vi.
Embed _content_vi/tour alongside _content/tour. TourOnly() now returns an overlayFS (vi → en) so the tour serves Vietnamese content (title: "Hành trình khám phá Go") while missing files fall back to English.
Awesome-list style index linking to every page on go.dev, organized for learners and practitioners. Includes tutorials, blog posts by category, language spec, Tour of Go lessons, and 67 talks (2010-2019).
Serve Vietnamese content
The tour article content was already translated, but the surrounding UI was still in English. This translates the remaining pieces: - values.js: module titles and descriptions in the table of contents (Basics, Methods and interfaces, Generics, Concurrency), plus all i18n strings (button tooltips, status messages, navigation labels) - editor.html: translate the "Imports" button label to "Nhập khẩu" - app.css: translate CSS-generated "on"/"off" toggle labels to "bật"/"tắt" for the syntax and imports checkboxes The github-repo field in the feedback template now points to tamnd/golang-website-vi instead of golang/go.
Translate remaining Go Tour UI to Vietnamese
Merge golang/website master (up to 8b62ccf) and update Vietnamese translations to match: - blog/inliner.md: add Lasse Folger to the credits list - doc/effective_go.html: update intro paragraph (no longer "new language"), replace the old January 2022 note with a concise warning box at the top, change "item" to "variable" in new(T) description, and add the Go 1.26 new(expr) paragraph
Sync upstream and translate updated content
Sync upstream golang/website through 786360a
The Go 1.27 release post, one of the nine files the upstream sync added that had no Vietnamese at all. It is the first page translated through tamnd/godev-vn-translator end to end, so it also brings translations.json, the manifest that records what each translation was made from. Without it there is no way to tell a translation whose English has moved from one that is current, and after a sync the only safe reading is that all of them are stale. The first answer came back with 24 escaped backticks in it, so a code span inside a link read as two literal backslashes and a backtick on the page. Every other gate passed that file. The audit caught it, the piece it was in was sent back with the finding attached, and this is the repair.
_content_vi/blog: translate go1.27, and add translations.json
The file still said to open pull requests against tamnd/golang-website-vi, which has not been this repository's name since the rename, and that was the whole of it. What it should say is the thing that is easy to get wrong here: the overlay filesystem means a missing or wrong translation renders as English and returns 200, so browsing the site tells you nothing about coverage. The audit does. Everything else in here is the shape a translation has to keep, which is what the fourteen gates check.
AGENT.md: rewrite for the renamed repo and the audit
…rom (#12) The manifest answers one question, has the English moved since this translation was made, and until now it could not be asked of any file except the one page translated through the tool. Every other translation came back as having no record, which is 654 notices that say nothing except that the corpus is older than the manifest. The answer was in git. For each Vietnamese file, the commit that last touched it is when that translation was current, and the English at that commit is what it was made from. Written by godev backfill, which is tamnd/godev-vn-translator#12. The result is worth reading carefully. L13 goes from 654 findings to 129: 97 of those are the files whose Vietnamese is byte for byte the English, which are copies rather than translations and are deliberately left without a record, and 32 are real. Those 32 are stale, they refuse, and they are exactly the 32 files the upstream sync modified. Nothing else in the corpus is named. So the refusal count goes up, from 195 to 227. That is the point. The sync invalidated 32 translations in March and there was no way to say which ones. Now there is, by name. Route, model and prompt hash are empty on these records. A translation made by hand was not asked for under any instructions the tool knows, and an empty field says so without inventing a value.
This fork serves _content_vi over _content, so Content and TourOnly return Vietnamese and the test suite it inherited from upstream cannot pass. Two call sites are the reason, and neither of them goes through the server's -content flag: internal/tour/tour.go holds website.TourOnly() in a package level variable and asserts >A Tour of Go< cmd/golangorg/server_test.go reads website.Content() to check that every release notes page carries its release date Those tests are not wrong. What they test is the renderer, and the fork has to not break the renderer. Whether the translation is any good is a different question, and the answer to it is the fourteen quality gates in tamnd/godev-vn-translator, not a grep for an English string. So there is a switch. GODEV_CONTENT=en returns the embedded English and restores the whole upstream suite. The default is Vietnamese, because a deployment that forgets to set an environment variable should serve the site this fork exists to serve. overlay_test.go covers the overlay itself, which is the one piece of behaviour in this repository that has no test anywhere upstream, and covers the switch. It asserts that GODEV_CONTENT changes what comes back rather than pinning a Vietnamese sentence, because a test that pins prose fails the next time somebody improves the prose.
This repository had no workflows at all, which for a fork that is 680 hand checked files of translated content is the wrong number of them. Two jobs. The first is the ordinary one: vet, build, and upstream's tests under GODEV_CONTENT=en, plus the overlay's own tests with the overlay on. cmd/screentest is skipped because it dials Google Cloud Storage for screenshot baselines and cannot pass without credentials, which is true upstream too. The second runs the fourteen gates from tamnd/godev-vn-translator against this checkout: dropped links, invented heading anchors, edited code blocks, translated template actions, front matter that moved, and whether each translation still matches the English it was made from. It runs with -max 227, which is a ratchet and not a target. The corpus has 227 refusals today. A required check demanding zero would be red on the first pull request and every one after it, including the ones fixing the refusals, and a check that is always red is a check somebody turns off. This one fails at 228. The number comes down in the same diff as the work that lowers it, which makes it a reviewable claim about what the change did, and the day it reaches zero the flag comes off and the gate is the plain one. go.mod stays at upstream's go 1.26.0. That line is a minimum language version and says nothing about what CI runs, so raising it in the fork would buy nothing and would put a seventh file into the divergence, to be conflicted on at every upstream sync. The Go version that matters is the one in this workflow, and it is 1.27. Actions are at checkout@v7, setup-go@v7 and upload-artifact@v7.
Author
|
Opened by mistake against the wrong remote, this belongs on a fork. Sorry for the noise. |
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.
This repository has no CI. It is a fork of golang/website carrying 680 files of
Vietnamese translation in an overlay filesystem, and until now nothing checked
either half of that on a pull request.
The switch
go test ./...cannot pass here as inherited.ContentandTourOnlyreturnthe overlay, so they return Vietnamese, and two upstream test call sites read
them directly rather than going through the server's
-contentflag:internal/tour/tour.go:32holdswebsite.TourOnly()in a package level variable and asserts>A Tour of Go<cmd/golangorg/server_test.go:347readswebsite.Content()to check every release notes page carries its release dateThose tests are not wrong to assert English. What they test is the renderer,
and the fork has to not break the renderer. So
GODEV_CONTENT=enturns theoverlay off and gives them back the content they were written against. The
default stays Vietnamese, because a deployment that forgets an environment
variable should serve the site this fork exists to serve.
overlay_test.gocovers the overlay itself, which has no test anywhereupstream, and covers the switch. It asserts that
GODEV_CONTENTchanges whatcomes back rather than pinning a Vietnamese sentence, because a test that pins
prose fails the next time somebody improves the prose.
The workflow
Two jobs.
testruns vet, build, and the upstream suite underGODEV_CONTENT=en, thenthe overlay tests with the overlay on.
cmd/screentestis excluded: it dialsGoogle Cloud Storage for screenshot baselines and cannot pass without
credentials, upstream included.
auditchecks out tamnd/godev-vn-translator and runs the fourteen qualitygates against this tree. Dropped links, invented heading anchors, edited code
blocks, translated template actions, front matter that moved, and whether each
translation still matches the English it was made from. The report is uploaded
as an artifact whether the gate passes or fails, since a failure is exactly
when it is worth reading.
Why -max 227 and not zero
The corpus has 227 refusals today. A required check demanding zero is red on
the first pull request and every one after it, including the ones fixing the
refusals, and a check that is always red is a check somebody turns off.
So the gate is pinned at what the corpus has and fails at 228. It only ever
goes down. A pull request that fixes ten refusals lowers the pin by ten in the
same diff, which is a reviewable claim about what the change did, and when it
reaches zero the flag comes off.
go.mod
Left at upstream's
go 1.26.0. That line is a minimum language version andsays nothing about what CI runs. Raising it in the fork would buy nothing and
would put a seventh file into the divergence, to be conflicted on at every
upstream sync. The Go version that matters is the one in the workflow, and it
is 1.27.
Checked locally
GODEV_CONTENT=en go testover every package butcmd/screentestpasses on Go 1.27internal/tourfails with the overlay on and passes with it off, which is the whole point of the switchgodev audit -max 227exits 0 at exactly 227 and would exit 1 at 228