RFR: 8364103: Convert existing sprintf-chains to stringStream [v2]

[Kim Barrett]

On Thu, 4 Sep 2025 13:12:35 GMT, Paul Hübner <phubner at openjdk.org> wrote:

>> Hi all,
>> 
>> This PR refactors `javaClasses`' `print_stack_element_to_stream` to use a `stringStream` instead of manually maintaining a buffer with `os::snprintf` chains.
>> 
>> The JBS issue outlines to do this for all occurrences of `snprintf`-like chains. The majority of the calls to `os::snprintf`, `os::snprintf_checked`, `os::vsnprintf`, etc. occur just once. `LogTagSet::vwrite` is the only place to my knowledge where there remains a `vsnprintf` chain using manual buffer arithmetic. After consulting @jdksjolen, we decided that due to the
>> a) low-level nature of the code, and 
>> b) widespread usage of logging in the VM, 
>> this should probably warrant a larger discussion and/or separate issue. Personally, I feel like the performance & correctness risks of introducing a high-level abstraction here outweighs the benefits. 
>> 
>> I've run tests with JDK tiers 1-3 on macOS (AArch64, x64), Linux (AArch64, x64), and Windows (x64); all green.
>
> Paul Hübner has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains three commits:
> 
>  - Incorporate reviewer feedback.
>  - Merge branch 'master' into JDK-8364103
>  - Refactor stack trace printing to use stringStream.

Looks okay as is, but there is some optional tidying up that could be done.

src/hotspot/share/classfile/javaClasses.cpp line 2615:

> 2613:   }
> 2614: 
> 2615:   stringStream ss;

Consider moving the stringStream declaration up near the front of the function, and moving the print of
`klass_name` and `method_name` earlier, to just after they are set up. And consider moving the setup
of `source_file_name` down closer to where it's used. The early setups before printing were needed in
the old code as part of calculating the buffer length. But that's no longer a concern.

src/hotspot/share/classfile/javaClasses.cpp line 2648:

> 2646:       nmethod* nm = method->code();
> 2647:       if (WizardMode && nm != nullptr) {
> 2648:         ss.print("(nmethod " INTPTR_FORMAT ")", (intptr_t)nm);

pre-existing: Usual practice is to use `p2i(nm)` rather than a cast.

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

Marked as reviewed by kbarrett (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/26894#pullrequestreview-3186410195
PR Review Comment: https://git.openjdk.org/jdk/pull/26894#discussion_r2322968092
PR Review Comment: https://git.openjdk.org/jdk/pull/26894#discussion_r2322951743