Possible segfault from passing italic=logical() to font_info()

[MichaelChirico]

Updating {systemfonts} from 1.2.3 to 1.3.1 breaks our {ggforce} (currently at 0.5.0), e.g. segfaulting this example:

https://github.com/thomasp85/ggforce/blob/128001400bee8ae1a27c772b6d9677660db01778/man/geom_mark_circle.Rd#L262-L264

The entrypoint into {systemfonts} from {ggforce} is thus:

systemfonts::font_info("", logical(), logical(), 12, res = 300)

It looks like the update to recycle vectors in #138 is related, because rep_len(logical(), 1L) becomes NA, which doesn't appear handled, e.g. here passes the NA integer value on to other libraries' code:

systemfonts/src/font_matching.cpp

Line 233 in beb28a8

locate_systemfont(

systemfonts/src/font_matching.cpp

Line 60 in beb28a8

key.italic = italic;

I threw some agents at this and they came up with two different solutions -- I don't know enough about the package to judge whether either is better, or if something else is off and another approach would be preferred.

1. Check NA from some helpers in FontDescriptor.h, e.g.

https://github.com/r-lib/systemfonts/blob/beb28a842456b3cbc0adb7d55fde9cb04985764e/src/FontDescriptor.h#L83-L8

+#include <Rmath.h>
inline int italic_to_fixed(double italic) {
+ if (ISNA(italic)) return 0;
  return italic < 0.0 ? 0 : (italic > 1.0 ? FIXED_MOD : italic * FIXED_MOD);
}

2. Check length-0 inputs from R and replace them with defaults, e.g.

systemfonts/R/font_info.R

Line 87 in beb28a8

italic <- rep_len(italic, full_length)

+   if (!length(italic)) italic <- FALSE
    italic <- rep_len(italic, full_length)
    weight <- rep_len(weight, full_length)
    width <- rep_len(width, full_length)

(happy to share either as a PR if you'd like)

I also couldn't figure out why this is an issue for us and not CRAN, probably some version mismatch (we are using R 4.5.0, FreeType 2.13.3, and fontconfig 2.16.2) or a stricter compiler. I did notice an error (not a crash) from the same lines on a {ggforce} PR: https://github.com/thomasp85/ggforce/actions/runs/19674040486/job/56350141145?pr=352.

It's possible this is WAI but it looked off to me to be possibly passing NA to a native library.