[wildfly-dev] Automatically resolving expressions in the CLI

[Brian Stansberry]

On 7/8/14, 6:37 AM, Edward Wertz wrote:
>
>
> ----- Original Message -----
>> On 7/7/14, 12:36 AM, Edward Wertz wrote:
>>> I've been thinking about the UX considerations over the last week.
>>> Generally there seem to be 3 basic situations involved.
>>>
>>> 1) There can be expressions in the result and the CLI can, in some
>>> manner, successfully resolve those expressions. This is dependent
>>> on the context that the command/operation is executed in,
>>> domain/standalone mode and the location in the configuration tree,
>>> but seems solvable.
>>>
>>> 2) There can be expressions in the result and the CLI can't resolve
>>> those expressions. For example if they're looking at a domain
>>> profile there's nothing to resolve against.
>>>
>>> 3) There won't be expressions in the result. There are lots of
>>> places in the CLI that would never return an expression. Which
>>> means the user would never need the argument/param.
>>>
>>> I think it's ok to hide the argument/param in situation 3. If
>>> expressions won't show up in the results there's no reason to have
>>> an option to resolve them really. It seems confusing to include it
>>> here.
>>>
>>
>> I don't really follow this. For sure we won't include any
>> --resolve-expressions param on commands where it isn't relevant, like
>> 'cd' or something. If that's what you mean, that's fine. Otherwise I
>> don't see what situation you're referring to.
>>
>
> I'm thinking about this as it applies to the 'ls' command mostly, since it seems to be the broadest command and is available basically everywhere in the system tree. However, there are lots of locations in that tree where expressions simply would never exist aren't there? Would 'ls' on the root ever return an expression? That information is all version numbers and names and, I suspect, it's impossible to see an expression at that location. Thus no reason to have the ability to resolve one. I'm thinking it's ok to hide the argument in that situation. I'm not sure how widely this problem exists with other operations and commands though. ls might be somewhat unique in this regard since it's available throughout the entire CLI.
>

Ok, I don't think we should worry about this. IMHO it would be 
confusing, since it would cause the --resolve-expressions param to not 
appear in locations where the user would have no intuitive understanding 
as to why not.

My assumption here is if --resolve-expressions were set, attributes that 
didn't have expressions would be displayed as they are now. So if a user 
set the param where there were no actual expressions, there's no harm.

>>> The main question is what to do in situation 2, where there will be
>>> expressions in a result but it's impossible to resolve. Hiding the
>>> argument there seems like it could be confusing for people.
>>> They'll see an expression and if they know the argument is
>>> available elsewhere they'll wonder why they can't use it here.
>>> Frustration ensues. I think it would be better to have some type
>>> of error message explaining, somehow, that their location within
>>> the configuration tree doesn't allow for expressions to be
>>> resolved. I'm not sure what it should say though or how many of
>>> these situations exist. Any suggestions?
>>>
>>
>> There are actually 2 variants of 2). One is the expression can't be
>> resolved at all, and the other is it can be resolved, but the
>> resolution
>> is not meaningful because the resolutions is not occurring in a
>> meaningful process.
>>
>> For example, in a managed domain, /profile=x/subsystem=y:
>>
>> max-size="${com.user.thingy.max-size:10}
>>
>> That can always be resolved, but the resolution is meaningless except
>> on
>> a server at /host=a/server=1/subsystem=y.
>>
>> A another example would be:
>>
>> enabled="${java.net.preferIPv4Stack}"
>>
>> where the system property might be set on a Host Controller, so it
>> resolves, but again the value is meaningless because a server might
>> have
>> a different value for the system property.
>>
>> Because of this, simply trying to resolve the expression and handling
>> a
>> failure will not work. (Sounds like a bad approach anyway.) The tool
>> is
>> going to have to know in advance whether the resolution can be
>> performed. That would either have to be statically built into the
>> tool
>> (bad) or the management API is going to have to indicate this for
>> each
>> resource. I see the latter being done by either adding a param to the
>> API of read-resource-description for resources where it is allowed,
>> or
>> by creating a separate op registered only where allowed.
>>
>> Any error handling I want done client side, i.e. if you and Alexey
>> decide to add --resolve-expressions everywhere and then put out a
>> failure if it's not supported, I want the CLI to decide to fail based
>> on
>> the management metadata, not to count on the server side providing
>> some
>> expected failure if it's not supported.
>>
>
> I definitely agree. I didn't make that clear, but it was always my intention for it to be a pre-determined failure message. I lumped the 'meaningless' resolve in with 'no resolve', and thought the CLI should be able to determine based on the location in the tree whether a resolve makes any sense. If it doesn't, an error message would stop the command from executing and tell the user it's not possible to resolve expressions at this location.
>

Ok, good. For the CLI to determine if the command is present it will 
need to do a :read-resource-description call against the current 
resource address.

>>> The other question is whether there are situations where 1 and 2
>>> actually overlap. Where some expressions are resolvable, but
>>> others are not. I haven't been able to figure out if that's an
>>> actual problem yet.
>>>
>>> I'm going to implement the ability to hide the argument for the
>>> 'ls' command this week, which Alexey pointed me towards on Friday,
>>> and look into how to add a param to the server-side operations.
>>> Once I have both working successfully I'll try to create a
>>> comprehensive list of all the applicable situations I can figure
>>> out. No sense in doing that before I can get it actually working
>>> for both the CLI commands and server-side operations first.
>>>
>>
>> What's tricky is the subtree under host=*, excluding host=*/server=*.
>> Everything in a domain not under host=* is cannot support resolution,
>> and everything under host=*/server=* can.
>>
>> Places where this is an issue are:
>>
>> /host=*/system-property=*
>>
>> This resource is not relevant on the HC process; it drives the config
>> of
>> the server. So --resolve-expressions cannot be supported here.
>>
>> /host=*/path=*
>>
>> This one is a bit tricky, as the expression must be resolvable on the
>> HC, so --resolve-expressions could work, but it's possible that the
>> path
>> gets passed to the server and gets resolved differently there. But I
>> think that's an ok semantic.
>>
>> /host=*/interface=*
>>
>> Same as path.
>>
>> /host=*/server-config=*/system-property=*
>> /host=*/server-config=*/interface=*
>> /host=*/server-config=*/path=*
>>
>> None of these resources are relevant on the HC process; they drive
>> the
>> config of the server. So --resolve-expressions cannot be supported
>> here.
>> It can be supported on /host=*/server-config=* itself though.
>>
>>> Joe Wertz
>>>
>>>
>>>
>>> ----- Original Message -----
>>>>
>>>> On 06/27/2014 04:45 PM, Brian Stansberry wrote:
>>>>> On 6/27/14, 8:27 AM, Alexey Loubyansky wrote:
>>>>>> On 06/26/2014 05:31 PM, Brian Stansberry wrote:
>>>>>>> Thanks, Joe, for looking into this.
>>>>>>>
>>>>>>> I'm curious what you've done so far with your 'ls
>>>>>>> --resolve-expressions'
>>>>>>> work. Did you use the existing
>>>>>>> ':resolve-expression(expression=___)' low
>>>>>>> level operation to process any expressions found in the
>>>>>>> :read-resource
>>>>>>> response?
>>>>>>>
>>>>>>> There are a few aspects of this I'd like to explore.
>>>>>>>
>>>>>>> One is the UX one. Is allowing 'resolve-expressions' in some
>>>>>>> contexts
>>>>>>> and not others a good UX? Will users understand that? I'm
>>>>>>> ambivalent
>>>>>>> about that, and am interested in others' opinions.
>>>>>>>
>>>>>>> If it can work for a server and for anything under /host=*,
>>>>>>> then
>>>>>>> I'm
>>>>>>> ambivalent. Any restriction at all is unintuitive, but once the
>>>>>>> user
>>>>>>> learns that there is a restriction, that's a pretty
>>>>>>> understandable one.
>>>>>>> If it only works for a patchwork of stuff under /host=* then
>>>>>>> I'm
>>>>>>> real
>>>>>>> negative about it. An area of concern is
>>>>>>> /host=*/server-config=*,
>>>>>>> where
>>>>>>> an expression might be irrelevant to the host, only resolving
>>>>>>> correctly
>>>>>>> on the server that is created using that server-config. That
>>>>>>> will
>>>>>>> need
>>>>>>> careful examination.
>>>>>>>
>>>>>>> A second one is how this data would be displayed with 'ls'. A
>>>>>>> separate
>>>>>>> additional column? Or replacing the current data? The answer to
>>>>>>> this
>>>>>>> might impact how it would be implemented server side.
>>>>>>
>>>>>> Keep in mind that ls is an example. There are other commands
>>>>>> that
>>>>>> will
>>>>>> have to support this feature once it's implemented in one place.
>>>>>> Another
>>>>>> example is read-attribute command. The ability to resolve
>>>>>> expressions
>>>>>> elsewhere will be a natural expectation then.
>>>>>> So, it has to be thought of as a general features that can be
>>>>>> applied to
>>>>>> various cli commands.
>>>>>>
>>>>>
>>>>> Good point. Joe, we'd need a clear understanding of all the
>>>>> commands
>>>>> that would be affected.
>>>>
>>>> At this point, it's ls, read-attribute and commands handled by
>>>> GenericTypeOperationHandler (which means [xa-]data-source,
>>>> jms-topic,
>>>> -queue, -connection-factory, etc).
>>>>
>>>> The generic handler includes action read-resource (e.g. w/o other
>>>> optional arguments 'data-source read-resource --name=ExampleDS'),
>>>> which
>>>> is basically a formatted result of :read-resource.
>>>>
>>>> In general, it could be applied to any command displaying an
>>>> attribute
>>>> value to the user.
>>>>
>>>> Alexey
>>>>
>>>>>
>>>>>> IMO, the values returned should just be replaced with the
>>>>>> resolved
>>>>>> ones
>>>>>> for display. Some commands support --verbose argument, in which
>>>>>> case
>>>>>> additional info is displayed in columns, there we could include
>>>>>> the
>>>>>> original value.
>>>>>> The output of the cli commands in some cases is parsed by
>>>>>> scripts
>>>>>> or
>>>>>> other code, so keeping it simple will help there too.
>>>>>>
>>>>>>> The third aspect is the technical issue of how to make any
>>>>>>> 'resolve-expressions' param or CLI argument available in
>>>>>>> certain
>>>>>>> contexts and not in others. That's very likely solvable on the
>>>>>>> server
>>>>>>> side; not sure how difficult it would be in the CLI high-level
>>>>>>> command.
>>>>>>
>>>>>> Current tab-completion supports dependencies of command
>>>>>> arguments
>>>>>> and
>>>>>> their values on the current context (connection to the
>>>>>> controller,
>>>>>> standalone/domain mode, the presence of other arguments on the
>>>>>> line and
>>>>>> the values specified for them, etc). Technically, there
>>>>>> shouldn't
>>>>>> be an
>>>>>> issue.
>>>>>
>>>>> Ok, good.
>>>>>
>>>>>> I am more concerned about how intuitive that will look like for
>>>>>> the user
>>>>>> in various contexts.
>>>>>>
>>>>>
>>>>> Yes, I think the UX aspects are the more significant ones.
>>>>>
>>>>>> Alexey
>>>>>>
>>>>>>> FYI, for others reading this, offline Joe pointed out there's a
>>>>>>> related
>>>>>>> JIRA for this: https://issues.jboss.org/browse/WFLY-1069.
>>>>>>>
>>>>>>> On 6/26/14, 5:41 AM, Edward Wertz wrote:
>>>>>>>> I'm looking into whether it's possible to automatically
>>>>>>>> resolve
>>>>>>>> expressions when executing operations and commands in the CLI.
>>>>>>>>
>>>>>>>> >From my understanding, there are two variations of the
>>>>>>>>> problem.
>>>>>>>>
>>>>>>>>        * Operations are server-side processes that are accessed
>>>>>>>>        via ':' in the CLI and, currently, the CLI presents the
>>>>>>>>        results returned as-is to the users. ex:
>>>>>>>>        ':read-resource'
>>>>>>>>
>>>>>>>>        * Commands are processes that get manipulated by the CLI
>>>>>>>>        before getting presented to users. ex: 'ls'
>>>>>>>>
>>>>>>>> I've been experimenting with adding arguments to the CLI
>>>>>>>> commands, like 'ls --resolve-expressions', and gotten it
>>>>>>>> working for the standalone and domain side of things. However,
>>>>>>>> I can't control the scope of the argument, so it's available
>>>>>>>> in
>>>>>>>> situations that cannot accurately resolve expressions like the
>>>>>>>> 'profile=full' section of the domain tree. The results
>>>>>>>> wouldn't
>>>>>>>> be reliable.
>>>>>>>>
>>>>>>>> The same problem would apply to adding parameters to the
>>>>>>>> server-side operations. The scope of the operations themselves
>>>>>>>> can be controlled, but not their parameters. An execution like
>>>>>>>> ':read-resource(recursive=true resolve-expressions=true)'
>>>>>>>> can't
>>>>>>>> resolve expressions unless it's used against an actual server
>>>>>>>> or host, but the operation is available almost everywhere.
>>>>>>>> Again, the results wouldn't be reliable.
>>>>>>>>
>>>>>>>> I'm wondering if anyone can suggest a way to attack this
>>>>>>>> problem? There is already a
>>>>>>>> ':resolve-expression(expression=___)' operation, so users can
>>>>>>>> somewhat laboriously get the runtime values they want, but I
>>>>>>>> can't figure out a way to integrate the values into the
>>>>>>>> existing framework successfully. Other than creating entirely
>>>>>>>> new operations and commands, like 'ls-resolve' and
>>>>>>>> ':read-resource-resolve', which seems like an unsustainable
>>>>>>>> way
>>>>>>>> to solve the problem.
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>>
>>>>>>>> Joe Wertz


-- 
Brian Stansberry
Senior Principal Software Engineer
JBoss by Red Hat