Let a second call refine the config verbs instead of replacing it (#119)

[mstackhouse]

Closes #119.

Your argument for merging is right, and it is a cleaner case than #113 for exactly the reason you gave: the state being merged is clinify's own config, where "not specified" is unambiguous, so there is nothing to guess.

ct |> clin_row_height(body = 15.35, title = 11.4, footnote = 11.4, rule = "atleast", unit = "pt") |>
      clin_row_height(header_leading = 0.75)
# body/title/footnote retained, header_leading added

It was not one function

Checking the siblings before fixing the reported one:

verb	second call naming only one thing	before
clin_row_height()	title = 11.4	body lost — as reported
clin_header_pad()	rule_to_body = 6	above lost
clin_spanner_rule()	rows = 1	border lost (a dashed pen reverted to the default)
clin_table_align()	—	fine, it carries a single value

All three now fold only the arguments a call actually names into what was already there. Fixing just the reported one would have left two known silent-data-loss bugs of the same shape.

Two things the fix turns on

Arguments with a default cannot be recognised by being NULL. rule and border are always populated, so a naive is.null() merge would still silently reset them. The merge keys off missing().

missing() has to be captured before match.arg() runs — assigning to a formal clears its missing status. Without that the fix looked correct everywhere except that a second call still reset rule to "atleast". Caught it in verification, and there is a test pinning it:

exact <- clin_row_height(base, body = 15.35, rule = "exact")
clin_row_height(exact, title = 11.4)          # rule stays "exact"
clin_row_height(exact, title = 11.4, rule = "atleast")   # named, so it wins
clin_row_height(base, body = 15.35)           # first call, default "atleast"

On rows, and why merging is still flat

clin_header_pad() and clin_spanner_rule() both carry a rows field that qualifies the values. Merging is field-wise, so a later call naming rows replaces the earlier rows for the whole setting rather than layering a second instruction. Documented as such. Layering different values on different header rows is the per-row vector form above = c(18, 34) from #113, which is a better fit than two calls anyway — I did not want two ways to express the same thing with different semantics.

Verification

• 2,176 tests pass, zero snapshot changes.
• devtools::check() — 0 errors, 0 warnings, and the unable to verify current time NOTE, which is the no-network note this repo's cran-comments.md documents as local-only.
• Lints on the three changed files: 0, excluding a cross-file object_usage_linter false positive for the shared helper, which single-file lint() cannot resolve and R CMD check's own pass does not flag.
• Tests cover each verb's merge, a named value still winning, rule surviving and being overridable, and a bare clin_spanner_rule() still ruling while border = FALSE still does not.

One correction to my own verification: I initially asserted isFALSE() on border = FALSE and a probe printed FALSE where I expected TRUE, which I glossed over before the test caught it. border = FALSE is normalised into a zero-width fp_border rather than kept as a logical — a sensible representation, and the assertion now checks width == 0 / style == "none".

Note

#120 is being investigated separately. DESCRIPTION left at 0.4.0 with the entry under # clinify (development version).

🤖 Generated with Claude Code