MobileRead Forums - View Single Post - epubveri

Kayadelenium · 08-28-2026, 03:20 AM

epubveri 0.13.0 is out.

This release is almost entirely false positives — twenty-six of them.

Twenty-three are the same shape: one defect reported twice. Each message was true, but the count did not match epubcheck's, and to anyone running both tools side by side that is indistinguishable from an invented error. Most were a hand-written check of mine saying what a schema had already said — a missing <spine>, a manifest <item> with no media-type, a duplicate manifest id, nested <dfn>, and so on.

Three were invented errors outright, and these are the ones that could reach a real book:

- <iframe srcdoc> was rejected. It is valid, and five more HTML5 iframe attributes were missing along with it: loading, sandbox, allowfullscreen, allow, referrerpolicy.
- Seventeen printable ASCII characters in a hostname were errors — a comma, and ! $ & ' ( ) * + ; = ~ ^ | { } and a backtick. epubcheck accepts all of them. (The comma is the one epubcheck's own corpus names, in issue #1034: it records that it should probably flag it, and does not. Reporting it anyway was my divergence, not a finding.)
- A media overlay whose stylesheet simply does not define the active-class was reported. epubcheck only reports when the document has no CSS at all.

Honest scope: exactly one of the twenty-six moves a book on my own shelf of 405 (the Adobe page-map one). The rest are visible only if you diff the two tools. They were found by running epubcheck over its own test fixtures and comparing the two reports line by line instead of by id — the id-set diff is blind to every one of them, because the extra finding carries the same id at the same severity. Comparing counts rather than id sets has now produced thirty-five findings in two days, two of them things I was missing rather than inventing.

There is one breaking change, but it is Rust-library-only (an enum variant now carries the names of the attributes an element is missing). The CLI, the JSON output and the WASM package are unchanged, so the Sigil plugin and anything parsing the JSON needs no update.

Downloads (8 platforms): https://github.com/veripublica/epubv...es/tag/v0.13.0
Full changelog: https://github.com/veripublica/epubv...n/CHANGELOG.md

As always, reports of a wrong error on a valid book are the most useful thing anyone can send me.

08-28-2026, 03:20 AM	#259
Kayadelenium Addict Posts: 207 Karma: 363834 Join Date: Jul 2026 Location: Planet Earth Device: Kobo Forma	epubveri 0.13.0 is out. This release is almost entirely false positives — twenty-six of them. Twenty-three are the same shape: one defect reported twice. Each message was true, but the count did not match epubcheck's, and to anyone running both tools side by side that is indistinguishable from an invented error. Most were a hand-written check of mine saying what a schema had already said — a missing <spine>, a manifest <item> with no media-type, a duplicate manifest id, nested <dfn>, and so on. Three were invented errors outright, and these are the ones that could reach a real book: - <iframe srcdoc> was rejected. It is valid, and five more HTML5 iframe attributes were missing along with it: loading, sandbox, allowfullscreen, allow, referrerpolicy. - Seventeen printable ASCII characters in a hostname were errors — a comma, and ! $ & ' ( ) * + ; = ~ ^ \| { } and a backtick. epubcheck accepts all of them. (The comma is the one epubcheck's own corpus names, in issue #1034: it records that it should probably flag it, and does not. Reporting it anyway was my divergence, not a finding.) - A media overlay whose stylesheet simply does not define the active-class was reported. epubcheck only reports when the document has no CSS at all. Honest scope: exactly one of the twenty-six moves a book on my own shelf of 405 (the Adobe page-map one). The rest are visible only if you diff the two tools. They were found by running epubcheck over its own test fixtures and comparing the two reports line by line instead of by id — the id-set diff is blind to every one of them, because the extra finding carries the same id at the same severity. Comparing counts rather than id sets has now produced thirty-five findings in two days, two of them things I was missing rather than inventing. There is one breaking change, but it is Rust-library-only (an enum variant now carries the names of the attributes an element is missing). The CLI, the JSON output and the WASM package are unchanged, so the Sigil plugin and anything parsing the JSON needs no update. Downloads (8 platforms): https://github.com/veripublica/epubv...es/tag/v0.13.0 Full changelog: https://github.com/veripublica/epubv...n/CHANGELOG.md As always, reports of a wrong error on a valid book are the most useful thing anyone can send me.