Re: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
Sorry for the spam.
Digging deeper through the related links on MNG-2339, it's apparent
that the comment in the DefaultMavenProjectBuilder is a touch
misleading. The key issues relevant to where sysprops get used during
interpolation are:
MNG-2745
MNG-2651
It seems that environments that use maven programmatically (in 2.0.x?
really??) are running into collisions where other libraries are
injecting system properties that override values from the POM for the
purposes of interpolation. For this reason (and because we don't have
a concept of CLI properties separate from sysprops yet), the code in
the project builder was changed to prefer values from the POM over
sysprops.
-john
On Mar 27, 2008, at 1:06 PM, John Casey wrote:
BTW, I found this comment on line 981 of DefaultMavenProjectBuilder:
// [MNG-2339] ensure the system properties are still
interpolated for backwards compat, but the model values must win
I've checked that issue, and it looks like it was closed for this
release...so, not present in 2.0.8. Additionally, the doesn't seem
to say anything about which is supposed to win - model vs.
sysprops. IMO, it makes more sense for CLI properties to override
those in the model, since it follows the principle of local-most
wins that we employ in other parts of Maven, but I'm not sure I
know enough about the history of this issue.
Does anyone have another issue number that contributes more to this
discussion, that we could use to determine the correct course of
action here?
Thanks,
-john
On Mar 27, 2008, at 12:54 PM, John Casey wrote:
Hmm, I'll have to do some homework on this one, but yeah, it looks
like the interpolation changes I put in to get the path-
translation in place. I'll have to see if I can work up a test
case for this, and try to track down that original issue.
Let me get to work on it and I'll see how fast I can come up with
something.
-john
On Mar 27, 2008, at 8:09 AM, Brian E. Fox wrote:
Hrm. It's probably a good idea to use a different property, but we
should understand why this changed before going further. John, any
ideas?
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of
Olivier Lamy
Sent: Thursday, March 27, 2008 6:56 AM
To: Maven Developers List
Subject: Re: [pre vote take 3] 2.0.9-RC3
Hi,
Testing on corporate projects and build fine.
+1
I have just noticed a change (regression ?).
We have a corporate plugin. In the pom it's configured as this :
plugin
..
configuration
subject.. - ${version} ../subject
We use it with mvn blabla -Dversion=here a version.
The value has changed :
- with mvn 2.0.8 : the value from the cli is used.
- with this RC : the ${version} is replaced with the current
pom.version.
It's not a blocking issue because we can easily replace with :
subject.. - ${releaseVersion} ../subject and use mvn blabla
-DreleaseVersion=
But I hope there is no other side effect.
--
Olivier
2008/3/26, Brian E. Fox [EMAIL PROTECTED]:
We fixed the regressions identified last week with the plugin tools
and
reporting impl. The new 2.0.9 is staged at
http://people.apache.org/~brianf/staging-repository/org/apache/
maven/apa
che-maven/2.0.9-RC3/
You'll notice that this one has an RC qualifier attached to it.
Since
what I've actually been staging hasn't been for an official
vote, it
makes more sense to have actual deterministic numbers on them
instead
of
continuously rolling back and forth between .10 and .9.
The other significant reason it has a qualifier is that I want to
solicit feedback from the users list without potentially getting
multiple versions out there called 2.0.9. My new mantra for the
maven
release is no more regressions. To that end, what I intend to
do is
let the RC sit here for a day. If no one turns up anything new (it
should be good since this is really attempt #3), then I'll
email the
user list to solicit feedback. Naturally we'll probably get a
slew of
can you fix xyz but the only thing that we will consider at this
point
would be a regression from 2.0.8 to the current RC. If
something is
identified then we should consider fixing it and re-releasing
RC4. I
think that having the users more involved in testing the RCs is
the
only
way to really identify and eliminate regressions. If someone
identifies
a regression after the fact and didn't speak up or try it, well
that's
unfortunate but it'll have to wait.
The RC can sit with the users for 3 days. If nothing turns up,
then
I'll
restage with a final release tag and we can do a formal vote.
Assuming
this is all successful, then I'll document a more formal Core
release
procedure that we can follow going forward.
Here's the list of issues fixed in the latest RC:
Release Notes - Maven 2 - Version 2.0.9
** Bug
* [MNG-1412] - dependency sorting in
RE: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
CLI should win. There was an issue open that I wrote for that a while
ago. I think it's still open even.
-Original Message-
From: John Casey [mailto:[EMAIL PROTECTED]
Sent: Thursday, March 27, 2008 1:07 PM
To: Maven Developers List
Subject: CLI Properties vs. Model Properties (Was Re: [pre vote take 3]
2.0.9-RC3)
BTW, I found this comment on line 981 of DefaultMavenProjectBuilder:
// [MNG-2339] ensure the system properties are still
interpolated for backwards compat, but the model values must win
I've checked that issue, and it looks like it was closed for this
release...so, not present in 2.0.8. Additionally, the doesn't seem to
say anything about which is supposed to win - model vs. sysprops.
IMO, it makes more sense for CLI properties to override those in the
model, since it follows the principle of local-most wins that we
employ in other parts of Maven, but I'm not sure I know enough about
the history of this issue.
Does anyone have another issue number that contributes more to this
discussion, that we could use to determine the correct course of
action here?
Thanks,
-john
On Mar 27, 2008, at 12:54 PM, John Casey wrote:
Hmm, I'll have to do some homework on this one, but yeah, it looks
like the interpolation changes I put in to get the path-translation
in place. I'll have to see if I can work up a test case for this,
and try to track down that original issue.
Let me get to work on it and I'll see how fast I can come up with
something.
-john
On Mar 27, 2008, at 8:09 AM, Brian E. Fox wrote:
Hrm. It's probably a good idea to use a different property, but we
should understand why this changed before going further. John, any
ideas?
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of
Olivier Lamy
Sent: Thursday, March 27, 2008 6:56 AM
To: Maven Developers List
Subject: Re: [pre vote take 3] 2.0.9-RC3
Hi,
Testing on corporate projects and build fine.
+1
I have just noticed a change (regression ?).
We have a corporate plugin. In the pom it's configured as this :
plugin
..
configuration
subject.. - ${version} ../subject
We use it with mvn blabla -Dversion=here a version.
The value has changed :
- with mvn 2.0.8 : the value from the cli is used.
- with this RC : the ${version} is replaced with the current
pom.version.
It's not a blocking issue because we can easily replace with :
subject.. - ${releaseVersion} ../subject and use mvn blabla
-DreleaseVersion=
But I hope there is no other side effect.
--
Olivier
2008/3/26, Brian E. Fox [EMAIL PROTECTED]:
We fixed the regressions identified last week with the plugin tools
and
reporting impl. The new 2.0.9 is staged at
http://people.apache.org/~brianf/staging-repository/org/apache/
maven/apa
che-maven/2.0.9-RC3/
You'll notice that this one has an RC qualifier attached to it.
Since
what I've actually been staging hasn't been for an official
vote, it
makes more sense to have actual deterministic numbers on them
instead
of
continuously rolling back and forth between .10 and .9.
The other significant reason it has a qualifier is that I want to
solicit feedback from the users list without potentially getting
multiple versions out there called 2.0.9. My new mantra for the
maven
release is no more regressions. To that end, what I intend to
do is
let the RC sit here for a day. If no one turns up anything new (it
should be good since this is really attempt #3), then I'll email
the
user list to solicit feedback. Naturally we'll probably get a
slew of
can you fix xyz but the only thing that we will consider at this
point
would be a regression from 2.0.8 to the current RC. If something is
identified then we should consider fixing it and re-releasing
RC4. I
think that having the users more involved in testing the RCs is the
only
way to really identify and eliminate regressions. If someone
identifies
a regression after the fact and didn't speak up or try it, well
that's
unfortunate but it'll have to wait.
The RC can sit with the users for 3 days. If nothing turns up, then
I'll
restage with a final release tag and we can do a formal vote.
Assuming
this is all successful, then I'll document a more formal Core
release
procedure that we can follow going forward.
Here's the list of issues fixed in the latest RC:
Release Notes - Maven 2 - Version 2.0.9
** Bug
* [MNG-1412] - dependency sorting in classpath
* [MNG-1914] - Wrong url in error message when using a mirror
* [MNG-2123] - NullPointerException when a dependency uses
version
range and another uses an actual version incompatible with that
range
* [MNG-2145] - Plugins' dependencies are not always checked
* [MNG-2178] - incorrect M2_HOME guess in mvn.bat
* [MNG-2234
RE: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
We have to err on the side of not causing more regressions. If we want
to move in this direction, we should start deprecating the non
${project. Forms of the properties with big warnings in 2.0.9.
-Original Message-
From: John Casey [mailto:[EMAIL PROTECTED]
Sent: Thursday, March 27, 2008 1:16 PM
To: Maven Developers List
Subject: Re: CLI Properties vs. Model Properties (Was Re: [pre vote take
3] 2.0.9-RC3)
Sorry for the spam.
Digging deeper through the related links on MNG-2339, it's apparent
that the comment in the DefaultMavenProjectBuilder is a touch
misleading. The key issues relevant to where sysprops get used during
interpolation are:
MNG-2745
MNG-2651
It seems that environments that use maven programmatically (in 2.0.x?
really??) are running into collisions where other libraries are
injecting system properties that override values from the POM for the
purposes of interpolation. For this reason (and because we don't have
a concept of CLI properties separate from sysprops yet), the code in
the project builder was changed to prefer values from the POM over
sysprops.
-john
On Mar 27, 2008, at 1:06 PM, John Casey wrote:
BTW, I found this comment on line 981 of DefaultMavenProjectBuilder:
// [MNG-2339] ensure the system properties are still
interpolated for backwards compat, but the model values must win
I've checked that issue, and it looks like it was closed for this
release...so, not present in 2.0.8. Additionally, the doesn't seem
to say anything about which is supposed to win - model vs.
sysprops. IMO, it makes more sense for CLI properties to override
those in the model, since it follows the principle of local-most
wins that we employ in other parts of Maven, but I'm not sure I
know enough about the history of this issue.
Does anyone have another issue number that contributes more to this
discussion, that we could use to determine the correct course of
action here?
Thanks,
-john
On Mar 27, 2008, at 12:54 PM, John Casey wrote:
Hmm, I'll have to do some homework on this one, but yeah, it looks
like the interpolation changes I put in to get the path-
translation in place. I'll have to see if I can work up a test
case for this, and try to track down that original issue.
Let me get to work on it and I'll see how fast I can come up with
something.
-john
On Mar 27, 2008, at 8:09 AM, Brian E. Fox wrote:
Hrm. It's probably a good idea to use a different property, but we
should understand why this changed before going further. John, any
ideas?
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of
Olivier Lamy
Sent: Thursday, March 27, 2008 6:56 AM
To: Maven Developers List
Subject: Re: [pre vote take 3] 2.0.9-RC3
Hi,
Testing on corporate projects and build fine.
+1
I have just noticed a change (regression ?).
We have a corporate plugin. In the pom it's configured as this :
plugin
..
configuration
subject.. - ${version} ../subject
We use it with mvn blabla -Dversion=here a version.
The value has changed :
- with mvn 2.0.8 : the value from the cli is used.
- with this RC : the ${version} is replaced with the current
pom.version.
It's not a blocking issue because we can easily replace with :
subject.. - ${releaseVersion} ../subject and use mvn blabla
-DreleaseVersion=
But I hope there is no other side effect.
--
Olivier
2008/3/26, Brian E. Fox [EMAIL PROTECTED]:
We fixed the regressions identified last week with the plugin tools
and
reporting impl. The new 2.0.9 is staged at
http://people.apache.org/~brianf/staging-repository/org/apache/
maven/apa
che-maven/2.0.9-RC3/
You'll notice that this one has an RC qualifier attached to it.
Since
what I've actually been staging hasn't been for an official
vote, it
makes more sense to have actual deterministic numbers on them
instead
of
continuously rolling back and forth between .10 and .9.
The other significant reason it has a qualifier is that I want to
solicit feedback from the users list without potentially getting
multiple versions out there called 2.0.9. My new mantra for the
maven
release is no more regressions. To that end, what I intend to
do is
let the RC sit here for a day. If no one turns up anything new (it
should be good since this is really attempt #3), then I'll
email the
user list to solicit feedback. Naturally we'll probably get a
slew of
can you fix xyz but the only thing that we will consider at this
point
would be a regression from 2.0.8 to the current RC. If
something is
identified then we should consider fixing it and re-releasing
RC4. I
think that having the users more involved in testing the RCs is
the
only
way to really identify and eliminate regressions. If someone
identifies
a regression after the fact and didn't speak up
Re: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
I was incorrect; this is not a result of code I changed. I'll have to
take a look at the SVN annotation to find the commit that changed
this, but it looks like it may have been part of some work Jason was
doing. I'm looking into it now.
-john
On Mar 27, 2008, at 1:19 PM, Brian E. Fox wrote:
We have to err on the side of not causing more regressions. If we want
to move in this direction, we should start deprecating the non
${project. Forms of the properties with big warnings in 2.0.9.
-Original Message-
From: John Casey [mailto:[EMAIL PROTECTED]
Sent: Thursday, March 27, 2008 1:16 PM
To: Maven Developers List
Subject: Re: CLI Properties vs. Model Properties (Was Re: [pre vote
take
3] 2.0.9-RC3)
Sorry for the spam.
Digging deeper through the related links on MNG-2339, it's apparent
that the comment in the DefaultMavenProjectBuilder is a touch
misleading. The key issues relevant to where sysprops get used during
interpolation are:
MNG-2745
MNG-2651
It seems that environments that use maven programmatically (in 2.0.x?
really??) are running into collisions where other libraries are
injecting system properties that override values from the POM for the
purposes of interpolation. For this reason (and because we don't have
a concept of CLI properties separate from sysprops yet), the code in
the project builder was changed to prefer values from the POM over
sysprops.
-john
On Mar 27, 2008, at 1:06 PM, John Casey wrote:
BTW, I found this comment on line 981 of DefaultMavenProjectBuilder:
// [MNG-2339] ensure the system properties are still
interpolated for backwards compat, but the model values must win
I've checked that issue, and it looks like it was closed for this
release...so, not present in 2.0.8. Additionally, the doesn't seem
to say anything about which is supposed to win - model vs.
sysprops. IMO, it makes more sense for CLI properties to override
those in the model, since it follows the principle of local-most
wins that we employ in other parts of Maven, but I'm not sure I
know enough about the history of this issue.
Does anyone have another issue number that contributes more to this
discussion, that we could use to determine the correct course of
action here?
Thanks,
-john
On Mar 27, 2008, at 12:54 PM, John Casey wrote:
Hmm, I'll have to do some homework on this one, but yeah, it looks
like the interpolation changes I put in to get the path-
translation in place. I'll have to see if I can work up a test
case for this, and try to track down that original issue.
Let me get to work on it and I'll see how fast I can come up with
something.
-john
On Mar 27, 2008, at 8:09 AM, Brian E. Fox wrote:
Hrm. It's probably a good idea to use a different property, but we
should understand why this changed before going further. John, any
ideas?
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of
Olivier Lamy
Sent: Thursday, March 27, 2008 6:56 AM
To: Maven Developers List
Subject: Re: [pre vote take 3] 2.0.9-RC3
Hi,
Testing on corporate projects and build fine.
+1
I have just noticed a change (regression ?).
We have a corporate plugin. In the pom it's configured as this :
plugin
..
configuration
subject.. - ${version} ../subject
We use it with mvn blabla -Dversion=here a version.
The value has changed :
- with mvn 2.0.8 : the value from the cli is used.
- with this RC : the ${version} is replaced with the current
pom.version.
It's not a blocking issue because we can easily replace with :
subject.. - ${releaseVersion} ../subject and use mvn blabla
-DreleaseVersion=
But I hope there is no other side effect.
--
Olivier
2008/3/26, Brian E. Fox [EMAIL PROTECTED]:
We fixed the regressions identified last week with the plugin
tools
and
reporting impl. The new 2.0.9 is staged at
http://people.apache.org/~brianf/staging-repository/org/apache/
maven/apa
che-maven/2.0.9-RC3/
You'll notice that this one has an RC qualifier attached to it.
Since
what I've actually been staging hasn't been for an official
vote, it
makes more sense to have actual deterministic numbers on them
instead
of
continuously rolling back and forth between .10 and .9.
The other significant reason it has a qualifier is that I want to
solicit feedback from the users list without potentially getting
multiple versions out there called 2.0.9. My new mantra for the
maven
release is no more regressions. To that end, what I intend to
do is
let the RC sit here for a day. If no one turns up anything new
(it
should be good since this is really attempt #3), then I'll
email the
user list to solicit feedback. Naturally we'll probably get a
slew of
can you fix xyz but the only thing that we will consider at
this
point
would be a regression from 2.0.8 to the current RC. If
something is
identified then we should consider fixing it and re-releasing
RC4. I
think that having the users
Re: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
I changed it, and it was in http://jira.codehaus.org/browse/MNG-2339
as the comment says.
JDK 1.4 defines a system property for version that is 2.4.1 for
some reason, and it wreaks havoc on anything that uses ${version}, $
{project.version}, etc.
I can't see why overriding model values makes any sense from the
command line - that's not what Olivier wanted but rather a straight
substitution.
The only change I can think of here is to have previous behaviour from
${version} and make sure ${project.version} is unaffected, but that's
a significant change to the interpolator.
I think we're better off leaving this fix in with something in the
release notes.
- Brett
On 28/03/2008, at 4:49 AM, John Casey wrote:
I was incorrect; this is not a result of code I changed. I'll have
to take a look at the SVN annotation to find the commit that changed
this, but it looks like it may have been part of some work Jason was
doing. I'm looking into it now.
-john
On Mar 27, 2008, at 1:19 PM, Brian E. Fox wrote:
We have to err on the side of not causing more regressions. If we
want
to move in this direction, we should start deprecating the non
${project. Forms of the properties with big warnings in 2.0.9.
-Original Message-
From: John Casey [mailto:[EMAIL PROTECTED]
Sent: Thursday, March 27, 2008 1:16 PM
To: Maven Developers List
Subject: Re: CLI Properties vs. Model Properties (Was Re: [pre vote
take
3] 2.0.9-RC3)
Sorry for the spam.
Digging deeper through the related links on MNG-2339, it's apparent
that the comment in the DefaultMavenProjectBuilder is a touch
misleading. The key issues relevant to where sysprops get used during
interpolation are:
MNG-2745
MNG-2651
It seems that environments that use maven programmatically (in 2.0.x?
really??) are running into collisions where other libraries are
injecting system properties that override values from the POM for the
purposes of interpolation. For this reason (and because we don't have
a concept of CLI properties separate from sysprops yet), the code in
the project builder was changed to prefer values from the POM over
sysprops.
-john
On Mar 27, 2008, at 1:06 PM, John Casey wrote:
BTW, I found this comment on line 981 of DefaultMavenProjectBuilder:
// [MNG-2339] ensure the system properties are still
interpolated for backwards compat, but the model values must win
I've checked that issue, and it looks like it was closed for this
release...so, not present in 2.0.8. Additionally, the doesn't seem
to say anything about which is supposed to win - model vs.
sysprops. IMO, it makes more sense for CLI properties to override
those in the model, since it follows the principle of local-most
wins that we employ in other parts of Maven, but I'm not sure I
know enough about the history of this issue.
Does anyone have another issue number that contributes more to this
discussion, that we could use to determine the correct course of
action here?
Thanks,
-john
On Mar 27, 2008, at 12:54 PM, John Casey wrote:
Hmm, I'll have to do some homework on this one, but yeah, it looks
like the interpolation changes I put in to get the path-
translation in place. I'll have to see if I can work up a test
case for this, and try to track down that original issue.
Let me get to work on it and I'll see how fast I can come up with
something.
-john
On Mar 27, 2008, at 8:09 AM, Brian E. Fox wrote:
Hrm. It's probably a good idea to use a different property, but we
should understand why this changed before going further. John, any
ideas?
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of
Olivier Lamy
Sent: Thursday, March 27, 2008 6:56 AM
To: Maven Developers List
Subject: Re: [pre vote take 3] 2.0.9-RC3
Hi,
Testing on corporate projects and build fine.
+1
I have just noticed a change (regression ?).
We have a corporate plugin. In the pom it's configured as this :
plugin
..
configuration
subject.. - ${version} ../subject
We use it with mvn blabla -Dversion=here a version.
The value has changed :
- with mvn 2.0.8 : the value from the cli is used.
- with this RC : the ${version} is replaced with the current
pom.version.
It's not a blocking issue because we can easily replace with :
subject.. - ${releaseVersion} ../subject and use mvn blabla
-DreleaseVersion=
But I hope there is no other side effect.
--
Olivier
2008/3/26, Brian E. Fox [EMAIL PROTECTED]:
We fixed the regressions identified last week with the plugin
tools
and
reporting impl. The new 2.0.9 is staged at
http://people.apache.org/~brianf/staging-repository/org/apache/
maven/apa
che-maven/2.0.9-RC3/
You'll notice that this one has an RC qualifier attached to it.
Since
what I've actually been staging hasn't been for an official
vote, it
makes more sense to have actual deterministic numbers on them
instead
of
continuously rolling back and forth between .10 and .9
RE: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
I can't see why overriding model values makes any sense from the
command line - that's not what Olivier wanted but rather a straight
substitution.
You shouldn't be able to override model values for sure. I was saying
that if something is defined on the CLI for everything else, it should
take precedence.
The only change I can think of here is to have previous behaviour from
${version} and make sure ${project.version} is unaffected, but that's
a significant change to the interpolator.
That's not promising.
I think we're better off leaving this fix in with something in the
release notes.
I don't. We need to stop shoving incompatible changes down the user's
throats. It shouldn't always require reworking of your poms to upgrade
to the next maven version. That's annoying and wrong. If we do need to
make changes, and I agree that forward progress must be made that
sometimes breaks stuff, it should be deprecated first so people can get
a chance to fix their poms.
Based on the votes and comments in the issue, it's clear we can't just
revert it either...We need to fix this correctly.
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
Brian, what do you consider the correct fix? That CLI takes precedence over
POM properties? I was trying to glean through this chain to find out your
proposal.
Paul
On Thu, Mar 27, 2008 at 2:08 PM, Brian E. Fox [EMAIL PROTECTED]
wrote:
I can't see why overriding model values makes any sense from the
command line - that's not what Olivier wanted but rather a straight
substitution.
You shouldn't be able to override model values for sure. I was saying
that if something is defined on the CLI for everything else, it should
take precedence.
The only change I can think of here is to have previous behaviour from
${version} and make sure ${project.version} is unaffected, but that's
a significant change to the interpolator.
That's not promising.
I think we're better off leaving this fix in with something in the
release notes.
I don't. We need to stop shoving incompatible changes down the user's
throats. It shouldn't always require reworking of your poms to upgrade
to the next maven version. That's annoying and wrong. If we do need to
make changes, and I agree that forward progress must be made that
sometimes breaks stuff, it should be deprecated first so people can get
a chance to fix their poms.
Based on the votes and comments in the issue, it's clear we can't just
revert it either...We need to fix this correctly.
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
On 28/03/2008, at 6:37 AM, Paul Benedict wrote:
Brian, what do you consider the correct fix? That CLI takes
precedence over
POM properties? I was trying to glean through this chain to find out
your
proposal.
The most correct fix is:
a) -Dversion should still replace ${version}
b) -Dversion should not replace ${project.version}
c) ${version} should evaluate to ${project.version} if nothing else
specified (though this should be deprecated behaviour).
The same applies for every other pom property, not just version.
The problem with my fix is that (a) no longer holds, though reverting
would sacrifice (b).
On IRC, John said he has a potential fix that makes (a) hold, and
while (b) doesn't, it mitigates the main observed side effect which is
-Dversion coming in from the environment, rather than the command
line. This is the least risk, but it's also another new behaviour we
may not want to preserve.
Fixing (a) + (b) is possible, but also risks some other regression in
(c).
- Brett
--
Brett Porter
[EMAIL PROTECTED]
http://blogs.exist.com/bporter/
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
RE: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
We're actively discussing on IRC, but in my mind a correct fix is one
that fixes the root of the jira, which is that system properties where
hosing versions, and doesn't break more builds.
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
On Behalf Of Paul Benedict
Sent: Thursday, March 27, 2008 3:38 PM
To: Maven Developers List
Subject: Re: CLI Properties vs. Model Properties (Was Re: [pre vote take
3] 2.0.9-RC3)
Brian, what do you consider the correct fix? That CLI takes precedence
over
POM properties? I was trying to glean through this chain to find out
your
proposal.
Paul
On Thu, Mar 27, 2008 at 2:08 PM, Brian E. Fox [EMAIL PROTECTED]
wrote:
I can't see why overriding model values makes any sense from the
command line - that's not what Olivier wanted but rather a straight
substitution.
You shouldn't be able to override model values for sure. I was saying
that if something is defined on the CLI for everything else, it should
take precedence.
The only change I can think of here is to have previous behaviour
from
${version} and make sure ${project.version} is unaffected, but that's
a significant change to the interpolator.
That's not promising.
I think we're better off leaving this fix in with something in the
release notes.
I don't. We need to stop shoving incompatible changes down the user's
throats. It shouldn't always require reworking of your poms to upgrade
to the next maven version. That's annoying and wrong. If we do need to
make changes, and I agree that forward progress must be made that
sometimes breaks stuff, it should be deprecated first so people can
get
a chance to fix their poms.
Based on the votes and comments in the issue, it's clear we can't just
revert it either...We need to fix this correctly.
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: CLI Properties vs. Model Properties (Was Re: [pre vote take 3] 2.0.9-RC3)
There should be absolutely no system properties manipulation in the
bowels of Maven. Or envars, they both need to be trapped at the front-
end because we end up with little bits here and there. They should all
be grabbed from the CLI, the order of precedence determined and then
passed into Maven as properties. All the system properties and envar
need to be entirely contained to the CLI. That's the direction I've
been moving in. Even if we obey things like ${ENV.foo} that should be
a property not operating on envars directly in the core.
On 27-Mar-08, at 10:49 AM, John Casey wrote:
I was incorrect; this is not a result of code I changed. I'll have
to take a look at the SVN annotation to find the commit that changed
this, but it looks like it may have been part of some work Jason was
doing. I'm looking into it now.
-john
On Mar 27, 2008, at 1:19 PM, Brian E. Fox wrote:
We have to err on the side of not causing more regressions. If we
want
to move in this direction, we should start deprecating the non
${project. Forms of the properties with big warnings in 2.0.9.
-Original Message-
From: John Casey [mailto:[EMAIL PROTECTED]
Sent: Thursday, March 27, 2008 1:16 PM
To: Maven Developers List
Subject: Re: CLI Properties vs. Model Properties (Was Re: [pre vote
take
3] 2.0.9-RC3)
Sorry for the spam.
Digging deeper through the related links on MNG-2339, it's apparent
that the comment in the DefaultMavenProjectBuilder is a touch
misleading. The key issues relevant to where sysprops get used during
interpolation are:
MNG-2745
MNG-2651
It seems that environments that use maven programmatically (in 2.0.x?
really??) are running into collisions where other libraries are
injecting system properties that override values from the POM for the
purposes of interpolation. For this reason (and because we don't have
a concept of CLI properties separate from sysprops yet), the code in
the project builder was changed to prefer values from the POM over
sysprops.
-john
On Mar 27, 2008, at 1:06 PM, John Casey wrote:
BTW, I found this comment on line 981 of DefaultMavenProjectBuilder:
// [MNG-2339] ensure the system properties are still
interpolated for backwards compat, but the model values must win
I've checked that issue, and it looks like it was closed for this
release...so, not present in 2.0.8. Additionally, the doesn't seem
to say anything about which is supposed to win - model vs.
sysprops. IMO, it makes more sense for CLI properties to override
those in the model, since it follows the principle of local-most
wins that we employ in other parts of Maven, but I'm not sure I
know enough about the history of this issue.
Does anyone have another issue number that contributes more to this
discussion, that we could use to determine the correct course of
action here?
Thanks,
-john
On Mar 27, 2008, at 12:54 PM, John Casey wrote:
Hmm, I'll have to do some homework on this one, but yeah, it looks
like the interpolation changes I put in to get the path-
translation in place. I'll have to see if I can work up a test
case for this, and try to track down that original issue.
Let me get to work on it and I'll see how fast I can come up with
something.
-john
On Mar 27, 2008, at 8:09 AM, Brian E. Fox wrote:
Hrm. It's probably a good idea to use a different property, but we
should understand why this changed before going further. John, any
ideas?
-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of
Olivier Lamy
Sent: Thursday, March 27, 2008 6:56 AM
To: Maven Developers List
Subject: Re: [pre vote take 3] 2.0.9-RC3
Hi,
Testing on corporate projects and build fine.
+1
I have just noticed a change (regression ?).
We have a corporate plugin. In the pom it's configured as this :
plugin
..
configuration
subject.. - ${version} ../subject
We use it with mvn blabla -Dversion=here a version.
The value has changed :
- with mvn 2.0.8 : the value from the cli is used.
- with this RC : the ${version} is replaced with the current
pom.version.
It's not a blocking issue because we can easily replace with :
subject.. - ${releaseVersion} ../subject and use mvn blabla
-DreleaseVersion=
But I hope there is no other side effect.
--
Olivier
2008/3/26, Brian E. Fox [EMAIL PROTECTED]:
We fixed the regressions identified last week with the plugin
tools
and
reporting impl. The new 2.0.9 is staged at
http://people.apache.org/~brianf/staging-repository/org/apache/
maven/apa
che-maven/2.0.9-RC3/
You'll notice that this one has an RC qualifier attached to it.
Since
what I've actually been staging hasn't been for an official
vote, it
makes more sense to have actual deterministic numbers on them
instead
of
continuously rolling back and forth between .10 and .9.
The other significant reason it has a qualifier is that I want to
solicit feedback from the users list without potentially getting
multiple
