Is this an example of abusing user-defined functions?

[paperchalice]

When I tried to register custom function by ICU4C, I found it is possible to implement function format like below:

example: {|placeholder: {$placeholder}| :format placeholder=|some text|}

It can produce example: placeholder: some text
Am I abusing this feature?

[eemeli]

This looks like an ICU4C bug that should be reported in its issue tracker: https://unicode-org.atlassian.net/jira/

Edit: Maybe I misunderstood? Is the :format operand value (the literal placeholder: {$placeholder}) being parsed separately within the function call? If so, this is technically valid but probably not great for the localizability of your messages.

[mihnita]

I don't think this is a bug in the ICU4C implementation.
Since :format is a custom function, it can do whatever the implementer of that function wants it to do.

My only problem is registering a custom function without a namespace.
The spec says (in spec/functions/README.md):

"Functions not defined by any version of this specification SHOULD use an implementation-defined or user-defined namespace."

With "SHOULD" interpreted as described in BCP 14 [RFC2119] and [RFC8174].

---

If this is an abuse or not, or if it is bad for l10n or not, depends on the intent of the function.

The literal placeholder (|placeholder: {$placeholder}|) is not localizable.
The argument (|some text|) might be localizable or not, it is function dependent.
And l10n would have to be informed about that.
I think this is one of the important pieces we are still missing. A machine-readable way to tell a l10n system what arguments are localizable and what is not.

Although I don't think it is technically incorrect, I cannot tell if such a function is a bad idea or not. Not until I can hear what the use case is.

[aphillips]

@paperchalice

Your example might be written differently, sorta like this:

.local $x = {|some text|}
.local $pattern = {|placeholder: {$x}|}
{{example: {$pattern :format placeholder=$x}}}

If your intention is to introduce a function that can format string fragments with placeholders in them, then that's, to @mihnita's point, something your function is permitted to do. (Please use a namespace)

It is probably a Bad Idea to have a function that does that for arbitrary text patterns, because MF2 is already designed to do your formatting. MF2's syntax was designed to avoid string concatenation masquerading as a function (it is what selectors are for), because it is difficult to do grammatical matching and agreement in such cases. Rendering your example into a sentence (instead of using subject:predicate form) might make clearer the problem:

.local $x = {|some text|}
.local $pattern = {|have {$x}|}
.match $numPeople
0   {{Nobody {$pattern :format placeholder=$x}}}
one {{{numPeople :number} person {$pattern :format placeholder=$x}}}
*   {{{numPeople :number} people {$pattern :format placeholder=$x}}}

... which produces ungrammatical results for the 0 and one cases and the translator can't fix it. Developers will use this feature to write messages that look okay in English but cannot be effectively translated (without fully rewriting the message).

This isn't to say that you could not have a function that uses a pattern string, though. Skeletons for dates work sort of like that, for example.

[paperchalice]

Thanks for replies. I would to split message to fit the mf2 model.

[aphillips]

@paperchalice

I would to split message to fit the mf2 model.

I don't think I understand your comment. Can you explain what :format is meant to do more clearly? Splitting your message up is exactly what MF2 allows you to avoid--and what you want it to avoid. But your example makes it hard to say what the right recommendation should be.

[paperchalice]

Sorry for the unclear context. I'm investigating translating clang diagnostic format string (very unfriendly to i18n) to mf2 string, clang's internal format strings can be nested, e.g.:

cannot find ROCm device library%select{| for %1| for ABI version %1%select{|, which requires ROCm %3 or higher}2}0;
provide its path via '--rocm-path' or '--rocm-device-lib-path', or pass '-nogpulib' to build without ROCm device library

%select{branch1|branch2|...}0 format, similar to mf2's matcher, can contain another %select format.
If use matcher here, the amount of cases will be 6, so I implement a :format function to handle the nested format string.
I think

[eemeli]

I don't see how more than four cases are needed?

.input {$sel0 :integer}
.input {$sel2 :integer}
.match $sel0 $sel2
0 * {{cannot find ROCm device library}}
1 * {{cannot find ROCm device library for {$arg1}}}
* 0 {{cannot find ROCm device library for ABI version {$arg1}}}
* * {{cannot find ROCm device library for ABI version {$arg1}, which requires ROCm {$arg3} or higher}}