Some possible improvements observed during .po revision

[rikivillalba]

• Here
  

  data.table/src/fwrite.c

  Lines 1119 to 1121 in b089b74

  	DTPRINT(_("Written %"PRId64" rows in %.3f secs using %d thread%s. MaxBuffUsed=%d%%\n"),
  	args.nrow, 1.0*(wallclock()-t0), nth, nth ==1 ? "" : "s", maxBuffUsedPC);
  	}

  
  better to avoid nth ==1 ? "" : "s", and use "thread/s" or similar.

• Here
  

  data.table/R/data.table.R

  Lines 895 to 896 in b089b74

  	plural_part <- sprintf(ngettext(length(tt), "The item in the 'by' or 'keyby' list is length %s.", "The items in the 'by' or 'keyby' list have lengths %s."), brackify(tt))
  	stopf("%s Each must be length %d; the same length as there are rows in x (after subsetting if i is provided).", plural_part, xnrow)

  
  I would've preferred

stop_msg = ngettext(
  length(tt), 
  "The item in the 'by' or 'keyby' list is length %s. Each must be length %d; the same length as there are rows in x (after subsetting if i is provided).", 
  "The items in the 'by' or 'keyby' list have lengths %s. Each must be length %d; the same length as there are rows in x (after subsetting if i is provided)."
)
stopf(stop_msg, brackify(tt),  xnrow)

used that plural_msg or stop_msg do not interfere with anything. It was not inmediately clear to me what "%s Each must be length %d... was meaning. (also we avoid <- right?)

[aitap]

Thank you very much for collecting these!

One more typo (extra ) at the end):

data.table/src/assign.c

Line 564 in b089b74

Rprintf(_("RHS for item %d has been duplicated because MAYBE_REFERENCED==%d MAYBE_SHARED==%d, but then is being plonked. length(values)==%d; length(cols)==%d)\n"),

I would also like to replace the use of ngettext for strings that don't format the count as a number because that prevents proper translation with plurals in some languages, but that can be a separate issue. For your example, that would be

stop_msg = if (length(tt) == 1) {
  gettext("The item in the 'by' or 'keyby' list is length %s. Each must be length %d; the same length as there are rows in x (after subsetting if i is provided).")
} else {
  gettext("The items in the 'by' or 'keyby' list have lengths %s. Each must be length %d; the same length as there are rows in x (after subsetting if i is provided).")
}
stopf(stop_msg, brackify(tt),  xnrow, domain = NA)

[venom1204]

hi @rikivillalba
can i work on it ?

[MichaelChirico]

@venom1204 thanks for offering. This will be handled as part of existing PR #6793

[MichaelChirico]

@aitap so sorry that I don't quite follow your issue with ngettext(), neither in the gettext thread nor in this example. In particular here the singular-plural difference is length vs lengths, that's a pretty normal singular-plural differentiation? Obviously I'm missing something.

[aitap]

What ngettext does best is not just distinguishing plural forms, but distinguishing the plural forms used for declension with a cardinal number in the same sentence. For example, Russian (together with Ukrainian, Belarusian, Serbian, Croatian) has three such declension forms. For %d elements, we get:

1. $n \in \{1, 21, 31, 41, 51, 61, 71, 81, 91,\dots\}$: «1 элемент», «21 элемент»: same declension for singular and some plural cases
2. $n \in \{2\dots{}4, 22\dots{}24, 32\dots{}34, 42\dots{}44, \dots\}$: «2 элемента»
3. $n \in \{5\dots{}20, 25\dots{}30, 35\dots40, \dots{}\}$: «5 элементов»

When translating a sentence that doesn't contain a cardinal number, those declension forms do not matter; a generic plural form must be used instead. "Multiple elements" ⇒ «несколько элементов», it is grammatically unimportant how many there were.

When R calls ngettext(n, "One element", "Multiple elements"), the function follows the Plural-Forms: header from the translation file to look up a declension form, not a generic singular/plural translation. In a language like Arabic, which has a special declension form for n==1, this means typing the singular translation once and the same generic plural translation five more times because Arabic has nplurals=6. In Russian, this means typing the generic plural form twice instead of declensions (2) and (3) above and then translating something like "one or multiple elements" instead of declension (1) because it includes n==1.

[rikivillalba]

Some grep'ed lines with ngettext() calls without cardinals, according to @aitap advice.
All the Pl_() calls i grep'ed so far have cardinals.

NOTE: potools::po_explain_plurals(<lang_code>) prints explanations for each lang.
NOTE 2: second example has a domain = NA parameter i don't know why

data.table/R/data.table.R

Lines 727 to 728 in f6dca2d

	else stopf(ngettext(sum(is.na(ansvals)), "column not found: %s", "columns not found: %s"),
	brackify(ansvars[is.na(ansvals)]), domain=NA)

R/data.table.R:712:            warningf(ngettext(sum(is.na(idx)), "column not removed because not found: %s", "columns not removed because not found: %s"),
R/data.table.R:727:          else stopf(ngettext(sum(is.na(ansvals)), "column not found: %s", "columns not found: %s"),
R/fmelt.R:27:    stopf(ngettext(length(idx), 'Pattern not found: [%s]', 'Patterns not found: [%s]'), brackify(p[idx]), domain=NA)

R/print.data.table.R:50:      ngettext(length(ixs), "Index: %s\n", "Indices: %s\n"),
R/setops.R:46:    stopf(ngettext(sum(found), "unsupported column type found in x or y: %s", "unsupported column types found in x or y: %s"),

R/utils.R:36-39
  stopf(ngettext(length(duplicate_names),
                 "%s has duplicated column name %s. Please remove or rename the duplicate and try again.",
                 "%s has duplicated column names %s. Please remove or rename the duplicates and try again."),
        table_name, brackify(duplicate_names), domain=NA)

[MichaelChirico]

Thanks, that's helpful. It does seem like a shortcoming of gettext(), e.g. suggesting 2 improvements:

(1) Russian (and similar) should have n==1 and n % 10 == 1 & n > 1 as different cases. I guess the downside is for "correct" usages of ngettext() like you're advocating, this will introduce duplication in a different way.
(2) .po format should support sequenced indexing like msgid[1:3] to avoid the need to copy-paste the message (and thereafter maintain the duplication). That would mitigate the concern of (1) too.

I've no hope that we'll see such changes anytime soon of course :)

[MichaelChirico]

I have a patch locally like:

diff --git a/R/translation.R b/R/translation.R
index fb398a85..cb447ceb 100644
--- a/R/translation.R
+++ b/R/translation.R
@@ -33,3 +33,13 @@ packageStartupMessagef = function(fmt, ..., appendLF=TRUE, class=NULL, domain="R
   # NB: packageStartupMessage() itself calls message(.packageStartupMessage(...))
   messagef(fmt, ..., appendLF=appendLF, class=c(class, "packageStartupMessage"), domain=domain)
 }
+
+# See discussion in #6787.
+# Not all `N==1 ? x : y`-type messages should generate msgid_plural markers, but it's still
+#   convenient to group them in an ngettext()-like fashion in the code. That lets us put the
+#   explanation for why ngettext() isn't used in just one place (here), as well as offering
+#   some readability to the knowing eye vs. a somewhat mysterious if/else branch in the main code
+#   that requires context switching from normal dev to fairly deep translation-aware-dev.
+ngettext_no_ordinal = function(n, msg1, msg2, domain="R-data.table") {
+  gettext(if (n == 1L) msg1 else msg2, domain=domain)
+}

But {potools} is not cooperating with it as a custom_translation_function so the .pot is not updating properly. This last part might be better to wait for the next release.