Backtestshugorun-20260922-141813#15078
New content command thinks there's a path conflict for a file that doesn't exist yet
The merged fix was judged better than our patch.
In this judgment, Patch A = our agent, Patch B = the merged fix.
The production fix is byte-for-byte identical in both patches: adding `draft: false` to the placeholder front matter in hugolib/content_factory.go. This defeats the cascade (`cascade.draft: true` from posts/drafts/_index.md), which was overriding the placeholder to draft:true and causing 'no page found ... conflicts with existing content'. So correctness, scope of the production change, and root-cause handling are a tie. The decision rests on test quality. Patch B adds a focused unit test (TestNewContentWithBuildCascade) in create/content_test.go that reuses the file's existing helpers (initFs, newTestCfg, testconfig.GetTestConfigs, readFileFromFs, and *qt.C which satisfies testing.TB), calls the function under test create.NewContent directly, and reproduces the issue's exact front matter including `build.render: never` and top-level `draft: true`. It fails before the fix (placeholder written) and passes after (real 'Trip to Puebla' content). It is idiomatic and consistent with neighboring tests. Patch A instead adds an integration testscript that deviates from the file's established inline-txtar convention and, more importantly, invokes `hugo new project issue-15078` — Hugo's scaffold subcommand is `hugo new site`, not `hugo new project` (Patch A's own next line correctly uses `hugo new content`), so that step likely fails as an unknown command, undermining confidence the test actually exercises the fix. Patch A also simplifies the reproducer, dropping `build.render: never` and the top-level `draft: true` from _index.md, making it a less faithful reproduction. Both patches touch exactly one test file plus the shared one-line prod change, so scope is equal; Patch B wins on a more reliable, idiomatic, and faithful test.
The judge saw only the two patches, labelled A and B in random order, with issue links and version numbers removed. The reasoning above is shown verbatim.
Solve
17m 48s43 turns
Wall-clock time in the container
Full test suite
No new failures
3,667 passed · 93 skipped · 3,773 total
Contamination audit
Unknown
See the audit notes for this run
Patch size
21vs 32 merged
Changed lines, ours vs merged
The issue
View on GitHubI run my personal blog using Hugo for about two years ago, and it has been great!, but there have been issues with how I create pages.
I have section called posts/drafts/ where I create drafts for posts before I move them to a posts/YEAR/MONTH directory where I can post them. The bug happens when I create the page for a post, like this:
<pre>
hugo new content posts/drafts/trip-to-puebla/index.md
</pre>
Running this command with the debug flag returns the following:
<pre>
INFO build: step process substep collect files 138 files_total 138 pagesources_total 71 resourcesources_total 67 duration 37.1674ms
INFO build: step process duration 38.2639ms
INFO build: step assemble duration 2.064063s
INFO build: step render pages 0 content 0 duration 548.6µs
INFO build: step render deferred count 0 duration 0s
INFO build: step postProcess duration 0s
INFO build: duration 2.1061069s
ERROR no page found for "C:\\...\\Documents\\projects\\kfcmanwebsite2\\content\\posts\\drafts\\trip-to-puebla\\index.md"; the target path conflicts with existing conten
</pre>
What's funny is that a page _is_ created, but it's a placeholder page that looks like this:
---
title: "Content Placeholder"
build:
render: never
list: never
publishResources: false
---
After consulting with ChatGPT (yes you can stone me now) and testing some things here's another stuff that cursed AI and I found:
- The name of the page still causes some issues, doesn't matter if its
trip-to-puebla/index.mdortrip-to-puebla. - Renaming the
draftsdirectory and rerunning the command with the new name still throws the error - This issue does not happen if I run it when creating something like
hugo new content posts/2026/07/foo/index.md
One final thing, I have a branch bundle on posts/drafts/_index.md that has the following front matter:
title: Drafts
build:
render: never
cascade:
draft: true
build:
render: link
weight: 9999
draft: true
This is so any of the pages inside posts/drafts don't get published.
What version of Hugo are you using (hugo version)?
<pre>
hugo v0.163.3-4d22555aebf458d5d150500c9ac4bee5b24cf0d3+extended windows/amd64 BuildDate=2026-06-18T16:18:24Z VendorInfo=gohugoio
</pre>
Does this issue reproduce with the latest release?
I'm using the latest version, so it _should_ be reproducible. I think. The repo for my hugo project can be found here, though it may be a little heavy.
The two patches
hugolib/content_factory.goindex 36cb29e63..6757c64e3 100644@@ -114,6 +114,7 @@ func (f ContentFactory) CreateContentPlaceHolder(filename string, force bool) (s // the paths correct. placeholder := `--- title: "Content Placeholder"+draft: false build: render: never list: nevertestscripts/commands/new_content.txtindex f432c8ace..94919836e 100644@@ -43,6 +43,26 @@ hugo new content P1.md ! hugo new content p1.md stderr 'the target path conflicts with existing content' +# Issue 15078++cd $WORK+hugo new project issue-15078+cd issue-15078+mkdir content/posts/drafts+cp $WORK/drafts-index.md content/posts/drafts/_index.md+hugo new content posts/drafts/trip-to-puebla/index.md+exists content/posts/drafts/trip-to-puebla/index.md+grep 'Trip to Puebla' content/posts/drafts/trip-to-puebla/index.md+! grep 'Content Placeholder' content/posts/drafts/trip-to-puebla/index.md++-- drafts-index.md --+---+title: Drafts+cascade:+ draft: true+ build:+ render: link+--- -- hugo.toml -- disableKinds = ['home','rss','section','sitemap','taxonomy','term'] [frontmatter]
create/content_test.goindex deafaed09..c8c3c1cdd 100644@@ -159,6 +159,37 @@ site RegularPages: {{ len site.RegularPages }} cContains(c, readFileFromFs(t, fs.Source, filepath.Join("content", "mypage.md")), `draft: true`) } +// See issue 15078.+func TestNewContentWithBuildCascade(t *testing.T) {+ t.Parallel()++ mm := afero.NewMemMapFs()+ c := qt.New(t)++ c.Assert(initFs(mm), qt.IsNil)+ c.Assert(mm.MkdirAll(filepath.Join("content", "posts", "drafts"), 0o755), qt.IsNil)+ c.Assert(afero.WriteFile(mm, filepath.Join("content", "posts", "drafts", "_index.md"), []byte(`---+title: Drafts+build:+ render: never+cascade:+ draft: true+ build:+ render: link+draft: true+---+`), 0o755), qt.IsNil)++ cfg, fs := newTestCfg(c, mm)+ conf := testconfig.GetTestConfigs(fs.Source, cfg)+ h, err := hugolib.NewHugoSites(deps.DepsCfg{Configs: conf, Fs: fs})+ c.Assert(err, qt.IsNil)++ const target = "posts/drafts/trip-to-puebla/index.md"+ c.Assert(create.NewContent(h, "", target, false), qt.IsNil)+ c.Assert(readFileFromFs(c, fs.Source, filepath.Join("content", target)), qt.Contains, `title: "Trip to Puebla"`)+}+ func initFs(fs afero.Fs) error { perm := os.FileMode(0o755) var err errorhugolib/content_factory.goindex 36cb29e63..6757c64e3 100644@@ -114,6 +114,7 @@ func (f ContentFactory) CreateContentPlaceHolder(filename string, force bool) (s // the paths correct. placeholder := `--- title: "Content Placeholder"+draft: false build: render: never list: never
What the agent reported
The working tree is clean and minimal: only my two intended changes plus the pre-existing AGENTS.md/CLAUDE.md deletions (present in the starting snapshot, unrelated to this fix). commands/new.go has no net change. Done.
Summary
What was wrong
hugo new content posts/drafts/trip-to-puebla/index.md failed with a false no page found for "…"; the target path conflicts with existing content error and left only a "Content Placeholder" file — but only when the target sat under a section whose branch-bundle _index.md had cascade: {draft: true}.
hugo new content writes a temporary placeholder content file, runs an internal non-draft build to resolve the page object (so the archetype can be applied), then overwrites the placeholder with the rendered archetype. The parent section's cascade: {draft: true} was marking the placeholder page as a draft; the non-draft build dropped it (content_map_page_assembler.go → site.shouldBuild), so HugoSites.GetContentPage returned nil and create/content.go reported the false conflict. Paths not under a draft cascade (e.g. posts/2026/07/foo/index.md) were unaffected — matching the report exactly.
What I changed
hugolib/content_factory.go— added a singledraft: falseline to the placeholder front matter inCreateContentPlaceHolder. A page's own front-matter value takes precedence over an inheritedcascadevalue (verified inhugolib/page__meta.go: front matter fillspageConfig.Paramsfirst; cascade only fills keys not already present), so the transient placeholder is no longer treated as a draft and survives the internal build. The placeholder is immediately overwritten by the applied archetype, sodraft: falsenever persists to the user's file.testscripts/commands/new_content.txt— added an "Issue 15078" regression scenario (+ a-- drafts-index.md --block) that creates a section withcascade: draft:true, runshugo new contentunder it, and asserts the real archetype output (Trip to Puebla) is present andContent Placeholderis not.
I intentionally did not enable a global BuildDrafts for this build (my first attempt did, and it broke the existing Issue 12786 scenario, which relies on a real draft leaf-bundle being dropped to detect a resource conflict). The surgical placeholder change preserves that behavior.
Test evidence
- Regression test failed before the fix with the exact reported error (
no page found … the target path conflicts with existing content). - After the fix:
TestCommands/new_contentpasses (all scenarios: 12599, 12602, 12786, 14112, 15078, archetypedir), and the fullcreatepackage (TestNewContentFromFile/*,TestNewContentFromDirSiteFunction) passes — confirming no regression in the shared placeholder path. - Reviewer confirmed the diff is minimal, matches the report, and cannot persist
draft: falseto user content.
Requirements checklist
- Fix the false "target path conflicts" error under a
draft: truecascade — implemented + tested (regression test fails before / passes after). - Behavior identical whether target is
trip-to-puebla/index.mdortrip-to-puebla— same code path (buildFile); the fix targets the shared placeholder, so both forms are covered. Not separately scripted, but the mechanism is path-form-independent. - Non-drafts case (
posts/2026/07/foo/index.md) must keep working — preserved; no cascade means no draft, andcreatetests + othernew_contentscenarios pass. - Don't break existing conflict detection (12602 dir conflict, 12786 leaf-bundle resource conflict, 14112 case collision) — verified; all pass, and I explicitly avoided the global-
BuildDraftsapproach that broke 12786.
All changes left uncommitted in the working tree.