bzr-git tests abort test run if dulwich mismatch

On Sun, 2010-02-28 at 20:13 +0000, Robert Collins wrote:
> Public bug reported:
>
> affects bzr-git
> done
>
> When I run
> 'bzr selftest plugin_info'
>
> bzr-git throws an exception and stops the test run.
>
> This is naughty: it should do one of:
> - fail to import (raise ImportError). This will be logged but the test
> run will continue
It does this but raises DependencyNotPresent rather than ImportError,
which is part of bzrlib.errors. Could bzr perhaps catch this error as
well?

Cheers,

Jelmer

On Sun, 2010-02-28 at 20:33 +0000, Jelmer Vernooij wrote:

> > This is naughty: it should do one of:
> > - fail to import (raise ImportError). This will be logged but the test
> > run will continue
> It does this but raises DependencyNotPresent rather than ImportError,
> which is part of bzrlib.errors. Could bzr perhaps catch this error as
> well?

I don't think the test infrastructure should do that: it won't be known
by other runners at all.

-Rob

----- Original message -----
> On Sun, 2010-02-28 at 20:33 +0000, Jelmer Vernooij wrote:
>
> > > This is naughty: it should do one of:
> > >  - fail to import (raise ImportError). This will be logged but the test
> > > run will continue
> > It does this but raises DependencyNotPresent rather than ImportError,
> > which is part of bzrlib.errors. Could bzr perhaps catch this error as
> > well?
>
> I don't think the test infrastructure should do that: it won't be known
> by other runners at all.
Can we then perhaps have DependencyNotPresent derive from ImportError ? Otherwise it's not really usable in plugins.

Cheers, jelmer

On Sun, 2010-02-28 at 22:15 +0000, Jelmer Vernooij wrote:
> ----- Original message -----
> > On Sun, 2010-02-28 at 20:33 +0000, Jelmer Vernooij wrote:
> >
> > > > This is naughty: it should do one of:
> > > > - fail to import (raise ImportError). This will be logged but the test
> > > > run will continue
> > > It does this but raises DependencyNotPresent rather than ImportError,
> > > which is part of bzrlib.errors. Could bzr perhaps catch this error as
> > > well?
> >
> > I don't think the test infrastructure should do that: it won't be known
> > by other runners at all.
> Can we then perhaps have DependencyNotPresent derive from ImportError ? Otherwise it's not really usable in plugins.

I don't understand why you need this in the test suite though.

Is it because you have

import dulwich

in a plugin source file? If so, if you wrap it in lazy_import, that
should work.

-Rob

On Sun, 2010-02-28 at 22:47 +0000, Robert Collins wrote:
> On Sun, 2010-02-28 at 22:15 +0000, Jelmer Vernooij wrote:
> > ----- Original message -----
> > > On Sun, 2010-02-28 at 20:33 +0000, Jelmer Vernooij wrote:
> > >
> > > > > This is naughty: it should do one of:
> > > > > - fail to import (raise ImportError). This will be logged but the test
> > > > > run will continue
> > > > It does this but raises DependencyNotPresent rather than ImportError,
> > > > which is part of bzrlib.errors. Could bzr perhaps catch this error as
> > > > well?
> > >
> > > I don't think the test infrastructure should do that: it won't be known
> > > by other runners at all.
> > Can we then perhaps have DependencyNotPresent derive from ImportError ? Otherwise it's not really usable in plugins.
>
> I don't understand why you need this in the test suite though.
>
> Is it because you have
>
> import dulwich
>
> in a plugin source file? If so, if you wrap it in lazy_import, that
> should work.
It seems a lot easier (and correct anyway) to just make
DependencyNotPresent derive from ImportError.

Also, lazy_import makes pyflakes useless.

Cheers,

Jelmer

On Mon, 2010-03-01 at 01:43 +0000, Jelmer Vernooij wrote:

> It seems a lot easier (and correct anyway) to just make
> DependencyNotPresent derive from ImportError.

DependencyNotPresent can occur outside of imports, so I don't think we
should change what it derives from (and multiple inheritance in
exceptions is unlikely to end well).

> Also, lazy_import makes pyflakes useless.

There is a patch for that.

-Rob

  status triaged
  importance low