On Tue, 22 Apr 2025 09:47:02 GMT, Kevin Walls <[email protected]> wrote:
>> src/jdk.hotspot.agent/linux/native/libsaproc/ps_proc.c line 450:
>>
>>> 448: if ( (ph = (struct ps_prochandle*) calloc(1, sizeof(struct
>>> ps_prochandle))) == NULL) {
>>> 449: snprintf(err_buf, err_buf_len, "can't allocate memory for
>>> ps_prochandle");
>>> 450: print_error("%s\n", err_buf);
>>
>> This might be one that would print it twice - the caller in
>> LinuxDebuggerLocal.cpp will print what's in err_buf.
>
> Looks like here in Pgrab the intent is to snprintf an error message in
> err_buf, and the caller prints it. So Pgrab should not need the print_error
> calls.
>
> Some functions which are called, like read_lib_info, may print their own
> specific error message, then Pgrab stores an error that "failed to read lib
> info".. These are reporting the same thing but I don't see that as a problem
> or duplication, it could be really useful to track where things fail.
>
> Yes it's a little annoying that Pgrab_core is different, it doesn't take an
> error_buf so needs to print_error about any problems.
ah, interesting. The native caller doesn't print the error message, it throws a
Java exception with the message in the exception string, and the Java code then
deals with the exception.
This is different on MacOS, where the error message is printed, and then an
exception with a generic message is thrown. I haven't checked what the Windows
implementation does here.
The Linux implementation with dynamic exception messages was introduced in
57d8a71115b8fc9a2eb2be876a396c474c207cf3.
I'll modify the Linux Pgrab to report all errors through the exception message,
and remove print_error from Pgrab.
Should I additionally modify the MacOS's
Java_sun_jvm_hotspot_debugger_bsd_BsdDebuggerLocal_attach0__I to report all
errors through the exception message and remove print_error, or is it OK if the
implementations behave differently?
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/24722#discussion_r2054467779
