I’m not exactly sure, but isn’t this just dialyzed way of telling you, that the else is missing?
Aside if that, there were issues with with and dialyzed in some elixir versions. So could you please tell us your elixir version?
Anyway, I think a repository to reproduce the issue would be nice.
Perhaps it might be enough if you just show the function that is called for the {:ok, 95} pattern. Perhaps dialzer thinks for some reason that that function can only return the error case?
The function is:
@spec validate_locale(Locale.locale_name() | LanguageTag.t()) ::
{:ok, String.t()} | {:error, {module(), String.t()}}
def validate_locale(%LanguageTag{cldr_locale_name: nil} = locale) do
{:error, Locale.locale_error(locale)}
end
def validate_locale(%LanguageTag{} = language_tag) do
{:ok, language_tag}
end
def validate_locale(locale_name) when is_binary(locale_name) do
locale =
locale_name
|> String.downcase
|> Cldr.Locale.locale_name_from_posix
|> do_validate_locale
case locale do
{:error, {Cldr.UnknownLocaleError, _}} -> {:error, Locale.locale_error(locale_name)}
{:error, reason} -> {:error, reason}
locale -> locale
end
end
def validate_locale(locale) do
{:error, Locale.locale_error(locale)}
end
Where locale -> locale is guaranteed to be {:ok, locale} -> {:ok, locale} (well, i believe its guaranteed).
I know 
I’ve been down the rabbit hole several times and didn’t want to burden the forum with more. But since you asked … (and this is really bugging me!):
for locale_name <- Cldr.Config.known_locale_names(config) do
language_tag =
Map.get(language_tags, locale_name)
|> Cldr.Locale.put_gettext_locale_name(config)
locale_name = String.downcase(locale_name)
defp do_validate_locale(unquote(locale_name)) do
{:ok, unquote(Macro.escape(language_tag))}
end
end
# It's not a well known locale so we need to
# parse and validate
defp do_validate_locale(locale) do
with {:ok, locale} <- Cldr.Locale.new(locale, unquote(backend)),
true <- !is_nil(locale.cldr_locale_name) do
{:ok, locale}
else
false -> {:error, Cldr.Locale.locale_error(locale)}
{:error, reason} -> {:error, reason}
end
end
Which is, I think, the end of the rabbit hole!
For anyone who got this far - thanks for any ideas! And congratulations on your fortitude!
What I’m wondering about is, that the pattern mentioned in the error is {:ok, 95}, But I can currently not find a path in the code that would generate that pattern anywhere…
Blockquote
Does this have anything to do with the typespec saying it returns {:ok, String.t()} when in this case it’s actually {:ok, integer()}?
Hmmmm, i was assuming the 95 was some kind of dialyzer reference not a literal. I cannot for the life of me see how a {:ok, 95} could be generated at all! Its just gets more puzzling.
Could be that do_validate_locale is returning {:ok, String.t()} or {:error, reason} where reason could be generated from locale_error
or other function not satisfying {:error, {module(), String.t()}}, ? which I think the type module() is an alias for atom()