The expected result is "μέλοσ"; see R6RS libraries section 1.2. However
instead Guile's result is "μέλος". Note that although Σ usually
downcases to σ, at the end of a string it's ς.
More precisely, it downcases to
σ if a letter follows and to
ς if not (being at the end of a string is a particular case). However, this is not actually always Greekly correct: the string "ΦΙΛΟΣ." with a period at the end downcases to "φιλος." if it is the word φίλος 'friend' (without its proper accent) at the end of a sentence, but as "φιλος." if it is an abbreviation for φιλοσοφία 'philosophy'. For this reason, R7RS does not require mapping to ς in this situation as R6RS does.
This test shows a
limitation of defining string-foldcase as simply (string-downcase
(string-upcase str)).
As explained in Unicode section 5.18, the foldcase mappings (in <
https://www.unicode.org/Public/UNIDATA/CaseFolding.txt>, the lines with status C and F) actually create a set of equivalence classes that are closed under {upper,lower,title}case mapping, and then choose a single character to represent each class. This is usually the unique lowercase character, but not always: in Cherokee it is the uppercase character, and in the set {Σ, σ, ς} it is
σ.
Good catch. I think there's even a worse example: dotless
and dotted I [1]. Here it seems even impossible to do
up- and downcase correctly without knowing the language
context.
Language-specific case mappings are explicitly out of Scheme's remit: they have to be performed by specialized libraries. There is an additional situation in Lithuanian dictionaries (but not running text): an "i" with a tone accent is represented as "i" + dot above + accent, like this: "i̇́". However, this dot above must be dropped when uppercasing, producing ordinary "Í".