Serialization warnings

[mbojan]

In statnet/ergm#641 we added neat progress bars. On a HPC which has, it seems, very stripped images on computing nodes, I'm getting locale-related warnings. Benign but annoying.

I thought its the progress bars, but apparently not.

Warning: strings not representable in native encoding will be translated to UTF-8
Warning: input string '
<f5><dc>
' cannot be translated to UTF-8, is it valid in 'UTF-8'?
Warning: input string '
<f5><dc>
' cannot be translated to UTF-8, is it valid in 'UTF-8'?
Warning: input string '
<f5><dc>
' cannot be translated to UTF-8, is it valid in 'UTF-8'?
Warning: input string '
<f5><dc>
' cannot be translated to UTF-8, is it valid in 'UTF-8'?
Burning in ======================>--------   74% |  ETA:  1s
Best valid proposal 'BDStratTNT' cannot take into account hint(s) 'triadic'.
Burning in ===============>---------------   49% |  ETA:  1s

[mbojan]

Claude writes:

Root cause (verified locally): both warnings come from R's ReadChar() in src/main/serialize.c, i.e. from unserializing a native-encoded, non-ASCII string that was serialized under a different native encoding. The string is the "crude sentinel" literal in statnet.common::match_names():

# src/statnet.common/R/misc.utilities.R:171-173
namesmatch <- if(partial) pmatch(names(v), names)
              else pmatch(paste(names(v), "\n\xf5\xdc\n"), paste(names, "\n\xf5\xdc\n"))

"\n\xf5\xdc\n" is invalid UTF-8 and unmarked (CE_NATIVE). When statnet.common is installed in a UTF-8 locale, the lazy-load DB records native encoding UTF-8; when the function is first lazy-loaded in a C-locale session, R tries to translate the string, fails, and warns — once for "strings not representable…" and once per copy of the literal (source + bytecode constant pool → 3). ergm triggers it on its first match_names() call (R/ergm.R:408/471, R/san.R, R/simulate.R, …), which lands right next to the burn-in bar.

Reproduction (no ergm needed):

LC_ALL=C Rscript -e 'options(warn=1); invisible(statnet.common::match_names)'