VBA Rubberducking

I don't mind that. what I do mind, is when the build fails and all I'm getting is a bunch of stupid CA warnings/"errors", and no clue whatsoever about the actual build error

this

oh yeah. That's annoying as hell

Comintern

9:06 PM

@MathieuGuindon Oh yeah, I think it was 15.7.8.

I'll have to check after work.

Mathieu Guindon

10 mins ago, by Mathieu Guindon

19 hours ago, by Comintern

Current is apparently 15.8.7, unless different editions have different versioning, but I doubt that.

this

would be nice to ensure that if it fails, stops building altogether

Comintern

Typo?

Hosch250

@MathieuGuindon No, that's right.

I just went to 15.8.8.

Mathieu Guindon

yeah

this

9:07 PM

don't keep on building because you have.... nothing.

mansellan

Speaking of releases, we releasing 2.3 soon?

Hosch250

It was a typo on my side.

this

@mansellan soon.

for some values of "soon".

Mathieu Guindon

@this 15 minutes later - build failed. would you like to run the tests off a useless outdated build that succeeded some time last week?

this

IKR?

Comintern

9:08 PM

I need a lull at work so I can get VS 2017 installed here.

Mathieu Guindon

@mansellan got anything pending for VB6?

Comintern

More like I need a lull at work so I can take a damned vacation...

Mathieu Guindon

I'm willing to ship 2.3.0 without block-AC, but SCP has to be there

this

But for me I think the worst thing is when I try to build and I'm still running RD. Deployment then get stuck for long time.

and if I remember to shut the host down, VS fails anyway

mansellan

@MathieuGuindon nothing that can't wait

Comintern

9:09 PM

We should probably fix the quick-fixes first.

Mathieu Guindon

gotta prepare a quick summary for these 1.4K commits..

Comintern

The inspection result bugs don't mangle code, but those do.

Mathieu Guindon

also there's a bunch of weird regression resolver issues that popped up recently

this

@MathieuGuindon "Hi. Stuff happened. Here's 2.3 version. Enjoy!"

Mathieu Guindon

lol

mansellan

9:11 PM

Gotta drive, bbl

Mathieu Guindon

later!

Comintern

Oh, there's the UI thing too.

#4361 - IIR that involves the parser

KySoto

i just thought up a funny

there are people out there that shouldn't be trusted to not get malware on a typewriter...

this

and hopefully we can fix the ghost process. That's the last (knock on wood) issue that prevents clean shutdown of RD. Who knows maybe we'll get lucky and we can close out the pUnk project. </pipedream>

Comintern

Wayne made that change to force a the TypeInfo stuff to run on the UI thread, but that means if the code doesn't compile, things break.

That might be underneath a handful of the weird resolver bugs too.

this

9:14 PM

Why should it need to be compiled?

I mean, we can't parse uncompilable code anyway?

Comintern

It doesn't.

this

Then I don't follow.

Comintern

The problem on 4361 is that it requests a reparse from a non-UI thread. That causes validation to fail:

Exception thrown: 'System.ArgumentException' in Rubberduck.VBEditor.dll
2018-09-20 19:15:25.0397;ERROR;Rubberduck.UI.Command.ReparseCommand;System.ArgumentException: Expected COM object, but validation failed.
   at Rubberduck.VBEditor.ComManagement.TypeLibs.TypeLibWrapper..ctor(IntPtr rawObjectPtr, Boolean isRefCountedInput) in

IIR there was a reparse requested in there too.

M.Doerner

That should really not be a problem.

Peter M Taylor

@Hosch250 got a link about Thermite as software right? Google tells me that Thermite is a self destructing hard disk!

M.Doerner

9:17 PM

In all places, the TypeLibWrapper is used, the execution should be pushed to the UI thread.

Comintern

Hmmm..., I could have sworn I wasn't on the UI thread there.

Hosch250

Thermite

Thermite () is a pyrotechnic composition of metal powder, which serves as fuel, and metal oxide. When ignited by heat, thermite undergoes an exothermic reduction-oxidation (redox) reaction. Most varieties are not explosive, but can create brief bursts of heat and high temperature in a small area. Its form of action is similar to that of other fuel-oxidizer mixtures, such as black powder. Thermites have diverse compositions. Fuels include aluminium, magnesium, titanium, zinc, silicon, and boron. Aluminium is common because of its high boiling point and low cost. Oxidizers include bismuth(III) oxide...

M.Doerner

However, that is something the caller must ensure.

this

Yeah, using UiDispatcher

Comintern

Sep 21 at 1:17, by Comintern

I'm not on the main thread.

M.Doerner

9:18 PM

Exactly

Hosch250

C-4 (explosive)

C-4 or Composition C-4 is a common variety of the plastic explosive family known as Composition C. A similar British plastic explosive, based on RDX but with different plasticizer than Composition C-4, is known as PE-4 (Plastic Explosive No. 4). C-4 is composed of explosives, plastic binder, plasticizer to make it malleable, and usually a marker or odorizing taggant chemical. C-4 has a texture similar to modelling clay and can be molded into any desired shape. C-4 is metastable and can be exploded only by the shock wave from a detonator or blasting cap. == Characteristics and uses == ==...

Peter M Taylor

@Hosch250 oh not a software got it. Thanks :)

Hosch250

No, blowing them up.

And burning the remains afterward.

(Or before--whichever you prefer.)

M.Doerner

Any call to the TypeLibWrapper should be put into a lambda and passed to one of the methods on the UiDispatcher.

this

Really, I think we need to start caching the data from the typelib

so that those data can be collected as part of the COM collecting

then the background components should not need to directly access the typelib API as much, right?

M.Doerner

9:26 PM

We are already caching the conditional compilation constants.

this

i'm not sure what the extract interface is trying to do there without seeing the code.

M.Doerner

Well, the test engine will still require it.

this

Yes, but that is actually OK since it'll have to be on UI

M.Doerner

I think it end by requesting a parse on a new task.

this

(I wrote UI originally but thinking about it, it's more that we need to avoid coding code that runs in background to access typelib API)

M.Doerner

9:28 PM

I think the test engine actually does the wrapping, IIRC.

this

with test engine it's not a problem because someone has to click the button to start the tests.

but if we run a background task and it may access the typelib stuff, that should not be done.

M.Doerner

Oh, the problem is not in the parsing process; it is the ReparseCommand.

this

but @Comintern said he wasn't on main thread anymore?

M.Doerner

It calls _typeLibApi.CompileProject(project).

this

hmm.

i guess to ensure that refactoring can be run, I suppose.

M.Doerner

9:32 PM

No, the refactoring requests a reparse in a new task it spinns off.

this

since it would mangle hte code if the code can't be compiled.

ah

M.Doerner

?

Comintern

Don't most of the refactorings do that?

this

so it's at the end of refactoring.

M.Doerner

I think most of them just request a reparse.

Comintern

9:32 PM

A refactor obviously invalidates at least one parse tree.

M.Doerner

They do not put the request into a new task.

That is useless anyway.

Comintern

Ah... gotcha.

M.Doerner

The ParseCoordinator parses in a new background task anyway.

this

hmm. Given that we have a setting to compile before parse, forcing a compile is potentially redundant.

M.Doerner

Still, the ReparseCommand should send the typeLibAPI stuff to the UiDispatcher.

The problem is in the code doing the compilation prior to the parse.

this

9:35 PM

I see.

M.Doerner

github.com/rubberduck-vba/Rubberduck/blob/…

this

Indeed.

That would be my bad

I think I naively assumed that it would be always on the UI thread since it would be from a button but that need not be the case.

M.Doerner

OK, what am I doing wrong? I have an interface that extends another interface and a concrete class implementing the extending interface. Then I have a call to a method of the extended interface on the concrete object where the argument passed is dynamic.

What I get is a runtime binder exception stating that the method is not on the interface.

Using the extended interface directly works.

this

Why can't you just cast the dynamic?

M.Doerner

Actually, what I am trying to do is split one method out of another interface.

The dynamic not my code there.

Specifically, I want to split Rewrite out of the IModuleRewriter into an IExecutableModuleRewriter in an attempt to change teh organization of rewriting.

Btw, I have no idea why the code uses dynamic in the SplitMultipleDeclarationsQuickFix.

this

9:51 PM

to be clear, Rewrite would no longer exist in the IModuleRewriter but now in IExecutableModuleRewriter?

M.Doerner

Yes

this

and the class SplitMultipleDeclarationsQuickFix would implement the IExecutableModuleRewriter, right?

M.Doerner

The problem is that it complains that Replace is not on IExecutableModuleRewriter.

No

That class just gets one from the parser state.

and uses it on a dynamic target, that might as well just be a ParserRuleContext.

this

TBH, I'm not sure you can do that without an explicit cast

M.Doerner

My idea is to hand out rewrite sessions instead of executable rewriters. The session hands out IModuleRewriters and saves the corresponding IExecutableModuleRewriters. Calling Rewrite on the session, does so on all rewriters requested.

this

9:55 PM

because from little I know, what dynamic may be at the runtime may not be what you want it to be, and must be cast if you wish to use a specific method that may not exist on the main object.

M.Doerner

Just to be clear, I am passing the dynamic thing to the rewriter, which is declared as IExecutableModuleRewriter, which extends IModuleRewriter, which has the method required.

Explicitly casting the rewriter to IModuleRewriter avoids the exception.

this

But that's the thing - dynamic only knows that it's IExecutableModuleRewriter

I don't think you get to access all possible interfaces with a dynamic and for that reason, an explicit cast is needed.

Example: stackoverflow.com/questions/7797207/…

M.Doerner

That is kind of strange. I always thought of extending an interface as making it castable to the other interface and including all its members.

this

I don't know - I have not worked with dynamic that much but from my brief exposure, I found that explicit casting them helps to avoid problems, and with pattern matching in the switch it is less of a problem than it was.

M.Doerner

Casting both arguments helped as well.

this

10:07 PM

do we actually need a dynamic, though?

M.Doerner

10:18 PM

Actually, pattern matching would work as well.

IvenBach

10:34 PM

@Hosch250 Thermite is cool stuff.

Comintern

@IvenBach ...and it's easy to make.

IvenBach

Shhhh... Don't let others know.

Comintern

Especially not today...

Duga

[rubberduck-vba/Rubberduck] comintern closed issue #3257: "Move closer to usage" Quick Fix breaks assignment of Word.Range using .duplicate

[rubberduck-vba/Rubberduck] comintern created comment on issue #3257: "Move closer to usage" Quick Fix breaks assignment of Word.Range using .duplicate

> I can't repro this on the current build.

[rubberduck-vba/Rubberduck] comintern project card. Enough said.

[rubberduck-vba/Rubberduck] comintern assigned comintern to issue #3258: "Move closer to usage" QuickFix breaks array declarations

this

11:19 PM

so I was investigating and I found out that for both UF and Access form, the VBComponent.Properties are inaccessible.

but they give out different error messages.

for Access.Form:

?application.VBE.ActiveVBProject.VBComponents(1).Properties.Count
---------------------------
Microsoft Visual Basic for Applications
---------------------------
Run-time error '35026':

Function or interface marked as restricted, or the function uses an Automation type not supported in Visual Basic
---------------------------
OK   Help
---------------------------

for the userform:

?application.VBE.ActiveVBProject.VBComponents(2).Properties.Count
---------------------------
Microsoft Visual Basic for Applications
---------------------------
Run-time error '-2147467259 (80004005)':

Automation error
Unspecified error
---------------------------
OK   Help
---------------------------

Comintern

.NET shouldn't care about the interface restriction, no?

this

Does anyone know if there's a chance that the errors are actually the same? It seems too concidental that the same command via VBIDE API would yield an error.

no, but we are getting it because we are going via VBIDE API.

in .NET, we are doing this: using (var properties = component.Properties)

where component is a IVBComponent

As you saw above, I was doing the near-equivalent from the VBE directly and I get the same error

Comintern

Well the HRESULT on the UF is "Unspecified failure". Very helpful there...

RubberDuck

lurk

Comintern

^ doing it wrong?

this

11:24 PM

^

@Comintern that's why I was wondering if MSForms was actually wrapping the original error or something silly like that

cos the error number & message has nothing to do with each other but they are both present if either are "open"

Comintern

I'm not getting any help from Google on RTE 35026.

this

but I can confirm that if either are open in design view, Properties works as expected

The problem, however, is that the original code has a dumb ol' try/catch block with no-op for catch

hhhmmm.

i better test that.

Comintern

Huh. Could it be that the OLE wrappers or compositor just aren't running when it isn't in design mode?

this

that's what I'm thinking originally but I need to see what it does w/ UF open in normal view

cos the report was that Access ghosts when we run the code w/ an open form. AFAIK, this hasn't been reported for UFs

Comintern

Heh. ?Err.Raise 35026 in the immediate window confuses the VBE.

this

11:32 PM

oh really?

Comintern

Yeah, it gives a compile error - argument not optional.

this

ah, no wonder

UF never happens. It's not a document

WTF?

ah, no you had ?

it's not a function.

Err.Raise 35026 just gives that application-defined or object-defined boilerplate message

Ok, so here's what I know so far....

Comintern

Yeah, that's what I got too. I was hoping to get luck with a description.

this

Once I hit VBComponent.Properties for that document, the no-op catch swallows it but at this point, Access is now ghosted.

We already have the properties in using, so it's being disposed

I even tried disposing the component (which I am not supposed to, BTW), didn't help

Comintern

Can you test in VBA while it's in design mode, or does that just fail?

this

11:37 PM

test what?

Comintern

Accessing properties to see if it ghosts.

That would narrow it down to RD hanging onto a resource as opposed to Access face planting.

this

i can quit just fine

so VBE somehow know enough to release whatever

or maybe it's not actually releasing

Comintern

Right. It probably just spins down the reference count.

this

but if that were the case, my silly component.Dispose() should have worked, no?

Comintern

Oh wait - it is the COM server. It can go down whenever the hell it wants.

this

11:40 PM

but once rD has parsed, it can't....

Comintern

Not if it causes Access itself to leak something internally.

this

hmm.

Comintern

Didn't we discuss at one point just not parsing when it's in design mode?

this

yes, what I wanted to avoid was adding host-specific stuff

because to know that information, you have to get it via Access API

rather than VBIDE API

Comintern

Right.

this

11:43 PM

btw, i tested differently

Public Sub derp()
On Error Resume Next
    Debug.Print Application.VBE.ActiveVBProject.VBComponents("Form_Form1Form").Properties.Count
End Sub

this still quits cleanly

Comintern

Interesting. That's using the compiled interface instead of the native one.

this

if you mean native = Access.Form, all Access.Form are exposed like that

e.g. VB project only see them as Form_***

Comintern

Is the property call trying to force it to instantiate the form? I.e., call CoCreate.

this

hmm.

Comintern

That would definitely leak.

Not entirely sure how to avoid it though.

The designer holds a running instance of the form, right?

this

11:47 PM

I would think so.

Comintern

Spy++ seems to think so.

this

if form isn't open: The expression you entered refers to an object that is closed or doesn't exist.

Comintern

It even has a window handle.

this

Access form itself, yeah it should

I can only actually get the count when the form is in design. If not opened or is in normal view, no go.

Comintern

Does the underlying class holding the hwnd change when it's in design mode?

this

11:48 PM

hmm, good question.

Comintern

I don't think the UF designers do, but I've never checked.

this

no

?forms(0).Hwnd
 2034886
 2034886

first was while it was in design; second while live

Comintern

Makes sense. I suspect OLE voodoo.

this

I think so. ActiveX stuff has additional interface for design time, no?

I'm pretty sure those interfaces shouldn't be available when it's not design time.

and for whatever reasons, this affects the VBComponent.Properties

Comintern

I don't think so - otherwise you wouldn't be able to, say, embed into a running spreadsheet or Word doc.

this

11:51 PM

One host-agnostic fix for this is to inspect if the interface is restricted and just not try to take the properties

but I don't think we expose that information via the declaration. AT least I coudln't find it.

Comintern

Isn't that interface always going to be restricted though?

this

nope, remember, it works when it's in design view

Comintern

That's a typelib flag, so I'm not sure how it could change.

this

only when it's in live view, then I can't access it.

oh... I would have to get it live

(i was assuming it got changed in between, since it's a dynamic class)

Comintern

Can you just call query interface on it, or does the member always resolve?

That might be implementation specific, but dollars to donuts they just built it with the ATL libs.

Although... Access.

this

11:54 PM

not sure what you mean there.

Comintern

ATL has a standard QI implemenation.

this

we are doing this:

var component = _projectsProvider.Component(qualifiedModuleName);
string parenthesizedName;
using (var properties = component.Properties)
using (var nameProperty = properties["Name"])

Comintern

It's like COM boilerplate in C++\MFC.

And component.Properties is the failing call, right?

this

the properties => IVBProperties, which wraps the COM Properties

yes

in the first attempt at least

next attempt, it will get past that, but then fail on properties["Name"]

Comintern

I was thinking of casting it to an unmanaged IDispatch or IUnknown and interrogating it about its Properties member.

this

11:56 PM