RFR: 8282773: Refactor parsing of integer VM options [v5]

[David Holmes]

On Mon, 14 Mar 2022 05:09:29 GMT, Ioi Lam <iklam at openjdk.org> wrote:

>> Background:
>> 
>> When a VM option is given an integer value (e.g., `-XX:ActiveProcessorCount=123`), we call the `set_numeric_flag` function in arguments.cpp. The old implementation always calls `strtoull()` to get an unsigned 64-bit number, and then tries to cast/truncate the result to the target type, performing sign conversion as needed.
>> 
>> This is cumbersome and buggy, because the target type (such as `size_t`) may have different bit-widths on different platforms. Also, manual conversion of unsigned to sign values can run into corner cases.
>> 
>> The fix:
>> 
>> - Use C++ templates and SFINAE to automatically choose the correct `parse_integer_impl()` function for parsing the input string. The VM should avoid doing cast and sign conversion as much as possible.
>> - Use a similar technique to write a comprehensive set of test cases to validate the implementation. (This has revealed several failures in the old implementation).
>> 
>> Notes on K/M/G/T suffix handling:
>> 
>> For input like `-XX:ActiveProcessorCount=2048m`, the old implementation checks for overflow with `(n * suffix / suffix) == n`. This works for unsigned types because the [C++ specification](https://en.cppreference.com/w/cpp/language/operator_arithmetic) allows unsigned arithmetic to overflow. However, overflowing a signed integer is UB. Therefore, I wrote a new function `multiply_by_1k()` to check for overflow without potentially undefined behavior. Thanks to @kimbarrett for helping me come up with the solution.
>> 
>> Tests:
>> 
>> tiers 1-2 passed. tiers 3-5 in progress.
>
> Ioi Lam has updated the pull request incrementally with one additional commit since the last revision:
> 
>   @kimbarrett comments -- avoid using "long" type and revert the implementation to 2a5e3e6df3e61b15126ca0582aada57955a25a07, and updated comments

Latest changes look good. I hadn't realized LP64_ONLY is defined on Windows when L is not 64-bit. :(

Thanks,
David

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

Marked as reviewed by dholmes (Reviewer).

PR: https://git.openjdk.java.net/jdk/pull/7763