Allow DateTimeFormatter constants to be used for the date pattern.

[jamezp]

Signed-off-by: James R. Perkins jperkins@redhat.com

[geoand]

Nice!

[geoand]

I recall seeing somewhere that ThreadLocal was used instead of a ConcurrentHashMap. Does that still exist? Is it a concern here?

[jamezp]

IIRC a ThreadLocal was used because we used the SimpleDateFormatter which is not thread safe. The DateTimeFormatter should be thread safe, so think it should work. TBH we might not even need to cache it. It's likely not invoked that often at all.

[geoand]

That's why I am asking, I don't know how often this is called :)

[dmlloyd]

I would advise against using reflection for this. It would tie the supported formats to the JDK in use, which would probably cause an issue at some point if and when we want to support (generally speaking) a format that only exists by name in a later JDK.

Also, the cache is probably overkill given that patterns are seldom created; I think a String-switch which resolves a name to the corresponding string or predefined constant is probably best. This allows us to control the namespace and also support formats that do not exist by name in the API, and also provides protection if such a format is introduced under a different name.

[jamezp]

Okay. That is a valid point. I was trying to think of a generic way to handle it, but it makes sense to keep it hard-coded.

[x80486]

Hello everyone! Is this one ready or is it waiting for something else?

[jamezp]

@x80486 It's on a temporary hold.

[dmlloyd]

What if instead of the enum, you did something along these lines in (half-Java, half-pseudocode):

private DateTimeFormatter resolve(String formatString) {
    if (formatString.length() > 1 && formatString.charAt(0) == '#') {
        formatString = formatString.substring(1);
        if (formatString.charAt(0) != '#') { // double-## means pass thru
            switch (formatString) {
                case "ISO_DATE": return DateTimeFormatter.ISO_DATE;
                // ... and one case for each known string that we want to allow ...
                default: return DateTimeFormatter.ofPattern("'<Unknown format " + formatString + ">'");
            }
        }
    }
    return DateTimeFormatter.ofPattern(formatString);
}

This keeps the code minimal and simple, is guaranteed to be compatible with future DateTimeFormatter utilizations of #, and is forward-compatible as well in case we want to support fancier things like putting the format name in mid-string or whatever.

[jamezp]

Do you mean the text format would be something like %d{#ISO_DATE}?

[dmlloyd]

Exactly. This way we never need to be concerned about syntactic conflicts between constant names and format strings.

[x80486]

On the other hand, that character really has no meaningful use from the user standpoint, with the exception of identifying, by convention, date/time pattern formats from constants — something you can achieve performing proper parsing.

I find it more confusing, in the sense that single characters in that particular string (e.g.: %date{ISO8601} |- %highlight(%5level) in %cyan(%logger{32}:%line) %magenta([%t - %m%n%ex) have a particular meaning.

Still it's a minor detail, but it amounts.

[jamezp]

@x80486 The main reason for the prefix character is most characters are reserved for a DateTimePattern. While ISO_DATE or ISO08601 is likely not going to bean issue, it could be a valid pattern at some point.