RFR 8160359: Improve jlink logging for cases when a plugin throws exception

[Seán Coffey]

I'm wondering if we could improve the information provided in the 
exception message. I don't think end users should have to turn on debug 
flags in all scenarios if the information could easily be inserted into 
the exception message.

e.g. 
src/jdk.jlink/share/classes/jdk/tools/jlink/internal/ImagePluginStack.java


>                            if (JlinkTask.DEBUG) {
> + System.err.println("IOException while reading resource: " + res.path());
>                                ex.printStackTrace();
>                            }
>                            throw new PluginException(ex);

Why not move this new message to the non-debug code ?

+ throw new PluginException("IOException while reading resource: " + 
res.path(), ex);

On the other hand, some messages might be deemed too verbose for 
exception message inclusion. Do end users mind how large an exception 
message is ?

e.g 
src/jdk.jlink/share/classes/jdk/tools/jlink/internal/PluginRepository.java
+                     System.err.println("Plugin " + plugin.getName() + 
" threw exception with config: " + config);

Including the config Map in an exception message might be deemed 
overkill for above.

Regards,
Sean.

On 09/11/16 03:43, Sundararajan Athijegannathan wrote:
> Mandy,
>
> The exceptions caught are propagated to caller - JlinkTask - which does
> print getMessage from exception and exits.  So, if a plugin throws
> illegal argument exception from "configure", jlink will print message
> and exit - as expected.
>
> Catching PluginException (and IllegalArgumentException in the case of
> configure) at precise location is a debugging aid. Plugin name, actual
> arguments map passed are dumped in addition to stack trace - only if
> system property is defined.
>
> -Sundar
>
>
> On 11/9/2016 2:12 AM, Mandy Chung wrote:
>> Sundar,
>>
>> Have you considered for plugins to throw different type of exceptions to differentiate if jlink should gracefully exit with no stack trace.  For example if it’s an user-error e.g. invalid option or duplicated entries then jlink can simply print the error message.
>>
>> For unexpected error such as IOException, the stack trace would be useful for troubleshooting.
>>
>> Mandy
>>
>> On Nov 8, 2016, at 7:51 AM, Claes Redestad <claes.redestad at oracle.com> wrote:
>>> Hi,
>>>
>>> I think this looks good as-is, but:
>>>
>>> Many times I've had the need to enable this kind of logging when
>>> building the JDK (as jlink plugins are now essential to creating the
>>> standard images), and having to hack the makefiles to add
>>> -J-Djlink.debug=true(?) to the right place seems a bit awkward.
>>>
>>> Should we simply make this added verbosity the default in the build?
>>>
>>> Thanks!
>>>
>>> /Claes
>>>
>>> On 2016-11-08 16:08, Sundararajan Athijegannathan wrote:
>>>> Please review for http://cr.openjdk.java.net/~sundar/8160359/webrev.00/
>>>> for https://bugs.openjdk.java.net/browse/JDK-8160359
>>>>
>>>> Thanks,
>>>>
>>>> -Sundar
>>>>