RFR: JDK-8266503: [UL] Make Decorations safely copy-able and reduce their size

[Thomas Stuefe]

On Thu, 6 May 2021 10:28:47 GMT, Yasumasa Suenaga <ysuenaga at openjdk.org> wrote:

>> This patch reduces the size of UL `LogDecorations` by about 85% (from 368 -> 56 bytes on Linux x64). This matters in the context of asynchronous logging in UL where we plan to keep a buffer containing log messages, including decorations, for asynchronous printing.
>> 
>> As a side effect, it makes the LogDecorations object safe to copy with trivial assignment constructors and operators (which it had not been before).
>> 
>> As another side effect, the 256-char-for-all-decorators limit has been removed with this patch.
>> 
>> What the patch does:
>> 
>> In LogDecorations, we resolve the values of the given decorators ("uptime", "tid" etc) and print them in human-readable format. Before this patch, the class LogDecorations stored the printed decorators in an internal (limited, fixed-sized) buffer. This is inefficient since this takes much more memory than storing the binary data before printing them.
>> 
>> So this patch separates the decorator value resolving from the printing. It stores the resolved values in binary form and only prints them when needed. Since a decorations object is only printed once this is fine. No need to cache the formatted text.
>> 
>> *Please Note that this patch includes the fix for JDK-8266536: "Provide a variant of os::iso8601_time which works with arbitrary timestamps" which is in a separate PR (https://github.com/openjdk/jdk/pull/3869) - so please ignore all the iso6801 stuff. It also contains a fix for the broken harfbuzz build.*
>> 
>> Testing:
>> - gtests, manually
>> - jtreg runtime/logging, manually
>> - SAP nightlies ran fine, but since then the tests changed, will run them again tonight
>
> src/java.desktop/share/native/libharfbuzz/hb-ot-shape-complex-use-machine.hh line 378:
> 
>> 376: {
>> 377:         machine_index_t (const Iter& it) : it (it) {}
>> 378:         machine_index_t (const machine_index_t& o) : hb_iter_with_fallback_t<machine_index_t<Iter>, typename Iter::item_t>(o), it (o.it) {}
> 
> This change does not seem to be for this PR...

Yes, this is a harfbuzz related build fix, see initial note. I'll remove it.

-------------

PR: https://git.openjdk.java.net/jdk/pull/3874