0004 · What a percentage becomes, and what @import becomes

Status: accepted, 2026-08-29. Decided by the maintainer. Constraint he set: resolve both without adding a keyword. Issue: #13

---

Three questions the emitter could not answer from the specification. None of the answers adds syntax: each is a fixed rule attached to a field or a construct that already exists.

1 · width = 80% becomes 0.80\linewidth

A percentage has to become a TeX length, and the reference changes the PDF. Compiled in a two-column document rather than recalled:

Inside	\linewidth	\textwidth	\columnwidth
figure — one column	229.5pt	469.0pt	229.5pt
figure* — spanning	469.0pt	469.0pt	229.5pt

\textwidth overflows a single-column float. \columnwidth under-fills a spanning one. \linewidth is the only reference correct in both, because it is the width of the line the float is actually set on.

Frequency in a corpus is the wrong metric here and was used at first: the author’s papers write \textwidth 124 times, \columnwidth 65 and \linewidth 49, and that says nothing about which is correct. A rule the compiler applies everywhere has to hold everywhere.

Naming a different reference

The percentage is a shorthand for the common case and its reference is fixed. An author who needs another one names it, and a length therefore admits a TeX length:

width = 80%                shorthand, 0.80\linewidth
width = 0.8\columnwidth     the reference named
width = \textwidth          no coefficient
width = 12cm                absolute

This was a hole rather than a trade. The first draft defined a length as a number plus one of six units, which left \columnwidth unreachable without abandoning the typed block for plain LaTeX — and inside a float spanning both columns, \linewidth is the full page width, so a column-width image there has no percentage form at all. The restriction was written without anyone asking whether it should hold. The maintainer asked.

A control word followed by { is a command taking an argument, not a length, and is rejected.

Checked before implementing: a backslash inside a field value does not disturb the block’s brace counting. The boundary is found at the same offset with and without one.

2 · height = 40% becomes 0.40\textheight

height is admitted as a figure field, with the same value kinds as width.

There is no \lineheight. TeX exposes no “space available in this float”, so no adaptive reference exists the way \linewidth does for width. \textheight — the page’s text height — is what the corpus uses whenever a relative height appears, and it is the only length with a defensible meaning here.

The asymmetry with width is real and it is not a design choice. It is what TeX offers.

3 · @import becomes \input, and could not have become \include

Not a preference. LaTeX settles it:

error: middle.tex:1: LaTeX Error: \include cannot be nested.

@import nests by definition — grammar.md §4 defines a document root as its root file plus the transitive closure of files reached by @import. A construct that nests cannot lower to a command that refuses to.

\input also imposes no page break, which \include does. Emitting a page break the author did not ask for would be injection under decisions/0001, so the two reasons agree.

What is still open: keepaspectratio

Giving \includegraphics both width and height stretches the image to fit unless keepaspectratio is also given. In the measured corpus, of 243 \includegraphics calls:

	Count
Give both width and height	17
…of those, also give keepaspectratio	15

So an author who gives both almost always means a bounding box rather than a stretch. Emitting keepaspectratio when a block gives both would follow that intent — and would be a second exception to no-injection, which decisions/0001 says needs its own record.

Not settled here. Until it is, a block giving both fields emits both and nothing else, and the image is stretched exactly as plain LaTeX would stretch it.