Find file paths built at runtime with sprintf()/paste()/paste0()/file.path()

A path is sometimes not written as a literal string but assembled from variables — read_csv(sprintf("%s/%s/x.csv", wd, wd_data)). This is fragile authoring (it silently breaks whenever the code runs from a different working directory than the author's own), and it also defeats code_file_refs()'s literal-string extraction: the "reference" it returns is really the RAW FORMAT STRING ("%s/%s/x.csv", placeholders and all), so a naive substring replacement of that text later corrupts the call — the surrounding sprintf(...) keeps its now-meaningless trailing arguments, , wd, wd_data), uselessly attached after the plan's target path was spliced in as if it were the whole first argument.

Usage

.repro_format_call_refs(code_text)

Arguments

code_text

the code text for a single file (character vector)

Value

a data frame with call_text (the full matched call, e.g. sprintf("%s/%s/x.csv", wd, wd_data)), fmt (the raw format string), resolved (the format string with resolvable placeholders substituted, or NA if any placeholder could not be resolved), and line (1-based line number the call starts on). Empty frame when the script has no such calls.

Details

This finds every such call in a script, and reports enough to fix it properly: the call's own full text (so the whole expression, not just the quoted substring, can be replaced), the raw format string, and — when every placeholder's corresponding argument is a simple variable resolvable to a literal string via .repro_simple_string_vars() — the format string with those values substituted in, which is what basename-matching should key on instead of the placeholder-laden original.

Only %s/%d placeholders (the common case in practice; a sprintf with a width/precision modifier like %05d is deliberately out of scope, since a zero-padded numeric ID is unlikely to appear in a bare data-file reference) are substituted; anything else in the format leaves that call unresolved (returned with resolved = NA), never guessed. The arguments after the format string are split on a plain comma, so an argument that is ITSELF a call containing a comma (sprintf("%s.csv", paste0(a, b))) is not split correctly either — this only means that argument then fails to resolve to any known variable (falling back to resolved = NA, the same safe unresolved outcome as any other unrecognised argument), never a wrong guess.