RFR: JDK-8318736: com/sun/jdi/JdwpOnThrowTest.java failed with "transport error 202: bind failed: Address already in use"

[Chris Plummer]

On Wed, 25 Oct 2023 16:53:18 GMT, Johannes Bechberger <jbechberger at openjdk.org> wrote:

>> test/lib/jdk/test/lib/JDWP.java line 60:
>> 
>>> 58:         if (parts.length != 2) {
>>> 59:             return null;
>>> 60:         }
>> 
>> This is a bit presumptuous in that it assumes that the first 2 token line of output is the line we are looking for. The reason `parseListenAddress()` tries to match a specific pattern is because sometimes tests are run in a way that causes the debuggee to produce some extra output (like logging output) and this output needs to be skipped.  At the very least you should be checking for a pattern that includes "dt_socket".
>> 
>> BTW, I tried using `printf` instead of `echo` so the debug agent could be made to generate a "Listening for transport..." line, in which case this new `parseLaunchEchoListenAddress()` would not be needed, but I could not get the argument processing right.  Something about exec() and printf just doesn't work right.
>
> Making too many not enough assumptions. The other method didn't make assumptions on the format of address and connection type, so I thought that I shouldn't too. I could create a temporary shell file that mimicks the other output though.

`parseListenAddress()` looks for a fairly specific format:

`listenRegexp = Pattern.compile("Listening for transport \\b(.+)\\b at address: \\b(.+)\\b");`

However, it does not assume the transport type, so maybe doing that here is not such a good idea. I don't like adding shell scripts to tests. We've been converting tests from shell scripts for years now because they tend to be problematic.

How about using something like launch="echo Listen Args: ". then you have a more specific string you can pattern match on:

`Listen Args: dt_socket 42305`

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16358#discussion_r1372079078