[Kenneth Cox]

Some of the tcltest constraint initialization code is
nontrivial. A caller may want to avoid running the
initialization code. In my case, running the
constraint initialization code hangs my application.

This patch uses [info exists] to test a constraint
first, and runs the code to set the constraint if it
doesn't already exist.

This patch is incomplete; it only handles the (stdio)
constraint. If you agree with this style of test-
first constraint setting, I can patch the other
constraints and submit them.

My example: I have a client app which opens an
exclusive connection to a server. When this app tries
to spawn itself (while setting the stdio constraint),
it deadlocks waiting for the exclusive connection to
the server. Whether this is good app design is
outside the scope of this patch!

Most of this patch is whitespace-only change.

• assigned_to: melissachawla --> dgp

• priority: 5 --> 3

Logged In: YES
user_id=80530

I haven't examine this patch, but I agree with the
sentiment.

I think the tcltest package still suffers from its legacy
as something that grew by accretion, and is something
mostly used exclusively in a single appliction. The
package/application boundary is not clean.

It would be better for the package to make functionality
available, and leave it to applications to actually exercise
it. In the case, the tcltest package should provide a means
for defining constraints, and it could even offer code that
queries the system and sets up some standard constraints.

But that code should only run in response to a command
executed by an application. It should not be part of
package initialization.

Tricky bit will be making this transition while supporting
compatibility -- perhaps some form of "auto-loading" of
the currently built-in constraints.

Not gonna get to this soon, but did want to comment.

• priority: 3 --> 7

Logged In: YES
user_id=80530

Here's a patch that converts tcltest to do
"lazy" initialization of test constraints.
The value of a built-in constraint is not
determined unless/until a [test] command
that depends on that constraint is evaluated.
This should address the problem.

The patch also deprecates the [initConstraintsHook]
because contraints are no longer initialized at
package load time by default. For compatibility, if
an [initConstraintsHook] has been defined, tcltest
reverts to the old behavior of initialization of
constraints at package load time. This should be
discouraged. The hook is completely unnecessary;
any tcltest user that wants to add a constraint can
do so directly by calling the [testConstraint] command.
No need for a hook at all.

This patch also cleans up the mess that was the
[limitConstraints] command. Some comments/documentation
indicated this command took a boolean value; some that
it took a list value. The existing code/tests did not
appear to do anything sensible. I replaced the whole
things with a consistent command that accepts a boolean
value indicating whether or not to limit running of [test]s
to those with constraints specified by the "-constraints"
command line option.

committing to tcltest 2.1 (Tcl 8.4a5)

• status: open --> closed-accepted