Re: exception handling

[Brett Porter <br...@apache.org>]

I'd planned to summarise what I did here tomorrow as its late...

I very carefully audited every exception in the codebase and ensured
that the message of each exception always captured any appropriate
message from the cause. There are no more "shoulder shrugs".

The root simply duplicated all of these (which looked silly), and is
easily obtained by the trace as you mention (with -e, not -X).

If you want to restore and:
a) you guarantee it doesn't duplicate the original message (which it did
even before I did the above audit)
b) that it actually gets the root cause (f.e., the root cause of
ArtifactResolutionException was coming back as "Could not get artifact
from any repositories" instead of the error of ResourceNotFoundException).

I'd appreciate testing on the error reporting. I think what we have now
is pretty good - can you find a specific example where it is not the case?

- Brett

John Casey wrote:

> Why on earth would you remove the root causes from all of these
> diagnosers?? Particularly the PuntDiagnoser?
>
> Often, the root cause is the only thing that makes any sense...the rest
> is just heaps of string-literal shoulder-shrugs that are too digested to
> be of any help. If we can't get at the root cause, especially when its
> hiding behind something like MojoExecutionException, or some otherwise
> undiagnosed exception (PuntDiagnoser), many problems will be a complete
> mystery. Not to mention the MNG issues this will reopen...
>

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
For additional commands, e-mail: dev-help@maven.apache.org

Re: exception handling

[John Casey <jd...@commonjava.org>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I only have 2 real problems here:

1. MojoExecutionException, which is meant to be a wrapper for some
unknown exception. This could be anything, and we REALLY cannot
guarantee that it's root cause is expressed in the top-level exception's
message, unless we're digging to root cause inside that exception (not
easy, given the restrictions on where you can call super(..)).

2. Unknown exceptions, which haven't been vetted as thoroughly as all
that. This could be from API changes to upstream components like SCM,
wagon, or plexus; or, it could be from new exceptions added to the maven
~ API without as thorough an examination as you apparently did.

In the first case, we simply need to push as far down as we can into the
exception stack and pull out the bottom exception. It would also
probably be worth pulling out the next-level (1 down from the top)
message and displaying that, just in case someone threw in useful
information at that level. Printing the error message thrown by the mojo
developer seems like we're putting too much faith in plugin developers
to produce accurate, thoughtful error messages. I'd say, pull
first-level and root-level error messages/types for these.

In the second case, we have an API expansion problem that hasn't had the
attention given to our current API. I agree that this isn't an ideal
situation, but if error diagnosis were in an ideal state, this diagnoser
could output "HELP" and it wouldn't matter; it'd never be seen. This is
the last line of defense for interpreting an unknown error...it should
print out the most useful information in any initial debugging effort:

a. top-level exception class
b. top-level exception message
c. root-level exception class
d. root-level exception message

Obviously, either of these cases will mean that determinedly dense
developers can cut the exception chain and throw a completely new
exception, swallowing the original exception and stack trace. However,
in these cases, we're left guessing at what the real problem is. Users
will be, too, and -X will be the first step in debugging. Why not simply
add a new layer, that attempts to help? If it doesn't help, we're right
back to -X, with no harm done.

Based on your reply, and my reply to that above, I'm putting these
changes back into those two diagnoser classes. If this is still a
problem, we can discuss it more in the morning (.au time).

- -j

Brett Porter wrote:
| I'd planned to summarise what I did here tomorrow as its late...
|
| I very carefully audited every exception in the codebase and ensured
| that the message of each exception always captured any appropriate
| message from the cause. There are no more "shoulder shrugs".
|
| The root simply duplicated all of these (which looked silly), and is
| easily obtained by the trace as you mention (with -e, not -X).
|
| If you want to restore and:
| a) you guarantee it doesn't duplicate the original message (which it did
| even before I did the above audit)
| b) that it actually gets the root cause (f.e., the root cause of
| ArtifactResolutionException was coming back as "Could not get artifact
| from any repositories" instead of the error of ResourceNotFoundException).
|
| I'd appreciate testing on the error reporting. I think what we have now
| is pretty good - can you find a specific example where it is not the case?
|
| - Brett
|
| John Casey wrote:
|
|
|>Why on earth would you remove the root causes from all of these
|>diagnosers?? Particularly the PuntDiagnoser?
|>
|>Often, the root cause is the only thing that makes any sense...the rest
|>is just heaps of string-literal shoulder-shrugs that are too digested to
|>be of any help. If we can't get at the root cause, especially when its
|>hiding behind something like MojoExecutionException, or some otherwise
|>undiagnosed exception (PuntDiagnoser), many problems will be a complete
|>mystery. Not to mention the MNG issues this will reopen...
|>
|
|
| ---------------------------------------------------------------------
| To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
| For additional commands, e-mail: dev-help@maven.apache.org
|
|
|
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (GNU/Linux)

iD8DBQFDTrVcK3h2CZwO/4URAqM+AKCdmaXzaeo//VvCZ8rnCW9ysVrHzQCdFH/P
kpZZLfQWaF9Bma1Ccc8Pv7E=
=pmwK
-----END PGP SIGNATURE-----

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
For additional commands, e-mail: dev-help@maven.apache.org

Re: exception handling

[John Casey <jd...@commonjava.org>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

What about component config errors in the plugin handling? I'll put it
back for artifact resolution too. I've done the MojoExecutionException
and the default diagnoser, and I'm checking that the main error message
doesn't include the root's message.

- -j

Brett Porter wrote:
| ... of course, I only did m2, not wagon.
|
| I'd suggest restoring the root cause for the
| ArtifactResolutionException, but make sure it grabs the IOException (I
| saw the 503 on the clean checkout), and don't display it if the main
| message contains that message.
|
| - Brett
|
| Brett Porter wrote:
|
|
|>I'd planned to summarise what I did here tomorrow as its late...
|>
|>I very carefully audited every exception in the codebase and ensured
|>that the message of each exception always captured any appropriate
|>message from the cause. There are no more "shoulder shrugs".
|>
|>The root simply duplicated all of these (which looked silly), and is
|>easily obtained by the trace as you mention (with -e, not -X).
|>
|>If you want to restore and:
|>a) you guarantee it doesn't duplicate the original message (which it did
|>even before I did the above audit)
|>b) that it actually gets the root cause (f.e., the root cause of
|>ArtifactResolutionException was coming back as "Could not get artifact
|
|>from any repositories" instead of the error of ResourceNotFoundException).
|
|>I'd appreciate testing on the error reporting. I think what we have now
|>is pretty good - can you find a specific example where it is not the case?
|>
|>- Brett
|>
|>John Casey wrote:
|>
|>
|>
|>
|>>Why on earth would you remove the root causes from all of these
|>>diagnosers?? Particularly the PuntDiagnoser?
|>>
|>>Often, the root cause is the only thing that makes any sense...the rest
|>>is just heaps of string-literal shoulder-shrugs that are too digested to
|>>be of any help. If we can't get at the root cause, especially when its
|>>hiding behind something like MojoExecutionException, or some otherwise
|>>undiagnosed exception (PuntDiagnoser), many problems will be a complete
|>>mystery. Not to mention the MNG issues this will reopen...
|>>
|>>
|>>
|>
|>---------------------------------------------------------------------
|>To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
|>For additional commands, e-mail: dev-help@maven.apache.org
|>
|>
|>
|
|
| ---------------------------------------------------------------------
| To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
| For additional commands, e-mail: dev-help@maven.apache.org
|
|
|
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (GNU/Linux)

iD8DBQFDTsTCK3h2CZwO/4URAhH+AJ4gJ1odYr9C14/cVIHbr5Pqy08U1QCfWWgT
Yh8CpyzVbulhqHqQCMhRLxc=
=rI1X
-----END PGP SIGNATURE-----

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
For additional commands, e-mail: dev-help@maven.apache.org

Re: exception handling

[Brett Porter <br...@apache.org>]

... of course, I only did m2, not wagon.

I'd suggest restoring the root cause for the
ArtifactResolutionException, but make sure it grabs the IOException (I
saw the 503 on the clean checkout), and don't display it if the main
message contains that message.

- Brett

Brett Porter wrote:

>I'd planned to summarise what I did here tomorrow as its late...
>
>I very carefully audited every exception in the codebase and ensured
>that the message of each exception always captured any appropriate
>message from the cause. There are no more "shoulder shrugs".
>
>The root simply duplicated all of these (which looked silly), and is
>easily obtained by the trace as you mention (with -e, not -X).
>
>If you want to restore and:
>a) you guarantee it doesn't duplicate the original message (which it did
>even before I did the above audit)
>b) that it actually gets the root cause (f.e., the root cause of
>ArtifactResolutionException was coming back as "Could not get artifact
>from any repositories" instead of the error of ResourceNotFoundException).
>
>I'd appreciate testing on the error reporting. I think what we have now
>is pretty good - can you find a specific example where it is not the case?
>
>- Brett
>
>John Casey wrote:
>
>  
>
>>Why on earth would you remove the root causes from all of these
>>diagnosers?? Particularly the PuntDiagnoser?
>>
>>Often, the root cause is the only thing that makes any sense...the rest
>>is just heaps of string-literal shoulder-shrugs that are too digested to
>>be of any help. If we can't get at the root cause, especially when its
>>hiding behind something like MojoExecutionException, or some otherwise
>>undiagnosed exception (PuntDiagnoser), many problems will be a complete
>>mystery. Not to mention the MNG issues this will reopen...
>>
>>    
>>
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
>For additional commands, e-mail: dev-help@maven.apache.org
>
>  
>

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
For additional commands, e-mail: dev-help@maven.apache.org