Eclipse Community Forums: TMF (Xtext)

Help

Home

Home » Modeling » TMF (Xtext) » Xtext hangs

Show: Today's Messages :: Show Polls :: Message Navigator

Xtext hangs [message #879696]

Thu, 31 May 2012 16:59

Oleg Bolshakov

Messages: 36
Registered: August 2010

Member

I have this grammar:

grammar org.xtext.example.mydsl1.MyDsl with org.eclipse.xtext.common.Terminals

generate myDsl "some link I can not post in the forum"

Model:
	elems+=MyRule*;
	
MyRule:
	'++' MyRule {MyRule.left = current} name = ID
	|'--' MyRule {MyRule.left = current} name = ID
        |'..' MyRule {MyRule.left = current} name = ID
;

The valid input I want to be is like this one:
"++ child1 ++ parent1 -- parent2"

Ans I want to get a tree like this:
name = parent2, left = {name = parent1, left = {name = child1, left = null}}

When I try to run MWE2, Xtext hangs just after "Registered GenModel".
when the third line in MyRule is deleted generating infrastructure works fine but the language does not work (I have many exceptions in runtime, the first is "The activator org.xtext.example.mydsl.ui.internal.MyDslActivator for bundle org.xtext.example.mydsl.ui is invalid") and the input "++ Hi" is not parsed correctly. What's wrong in the grammar?

Report message to a moderator

Re: Xtext hangs [message #879739 is a reply to message #879696]

Thu, 31 May 2012 18:52

Henrik Lindberg

Messages: 2509
Registered: July 2009

Senior Member

Well, there are several things that are a bit odd...

On 2012-31-05 18:59, Oleg Bolshakov wrote:
> I have this grammar:
> grammar org.xtext.example.mydsl1.MyDsl with
> org.eclipse.xtext.common.Terminals
>
> generate myDsl "some link I can not post in the forum"
>
> Model:
> elems+=MyRule*;
>
> MyRule:
> '++' MyRule {MyRule.left = current} name = ID
> |'--' MyRule {MyRule.left = current} name = ID
> |'..' MyRule {MyRule.left = current} name = ID
> ;
>
Given input ++a ++a is that two MyRule instances added to Model.elems or
one? (Now it is ambiguous)

You do not remember the operator ++, -- or .. there is no trace of those
in the resulting AST.

> The valid input I want to be is like this one:
> "++ child1 ++ parent1 -- parent2"
>
> Ans I want to get a tree like this:
> name = parent2, left = {name = parent1, left = {name = child1, left =
> null}}
>

How is the recursion supposed to end? You expect a null MyRule.left to
be at the end, but how would it ever get there since all the pre-ops
(++, --, ..) require that they are followed by a MyRule.

Did you perhaps want something like this:

Model : elems = CompoundMyRule?

CompoundMyRule : MyRule ({MyRule.left = current} MyRule)* ;
MyRule : op = ('++' |'--' |'..') name = ID ;

i.e. the last found MyRule is returned at the top of the result tree.
If there is only one, that MyRule has left == null.

> When I try to run MWE2, Xtext hangs just after "Registered GenModel".
> when the third line in MyRule is deleted generating infrastructure works
> fine but the language does not work (I have many exceptions in runtime,
> the first is "The activator
> org.xtext.example.mydsl.ui.internal.MyDslActivator for bundle
> org.xtext.example.mydsl.ui is invalid") and the input "++ Hi" is not
> parsed correctly. What's wrong in the grammar?

Not surprising, the grammar is highly ambiguous, and it looks like it
requires an infinite input to be able to parse a MyRule :).

- henrik

Report message to a moderator

Re: Xtext hangs [message #879759 is a reply to message #879696]

Thu, 31 May 2012 19:36

Sebastian Zarnekow

Messages: 3118
Registered: July 2009

Senior Member

Did you try to increase the amount of memory for the generator?

Regards,
Sebastian
--
Need professional support for Eclipse Modeling?
Go visit: http://xtext.itemis.com

Am 31.05.12 18:59, schrieb Oleg Bolshakov:
> I have this grammar:
> grammar org.xtext.example.mydsl1.MyDsl with
> org.eclipse.xtext.common.Terminals
>
> generate myDsl "some link I can not post in the forum"
>
> Model:
> elems+=MyRule*;
>
> MyRule:
> '++' MyRule {MyRule.left = current} name = ID
> |'--' MyRule {MyRule.left = current} name = ID
> |'..' MyRule {MyRule.left = current} name = ID
> ;
>
> The valid input I want to be is like this one:
> "++ child1 ++ parent1 -- parent2"
>
> Ans I want to get a tree like this:
> name = parent2, left = {name = parent1, left = {name = child1, left =
> null}}
>
> When I try to run MWE2, Xtext hangs just after "Registered GenModel".
> when the third line in MyRule is deleted generating infrastructure works
> fine but the language does not work (I have many exceptions in runtime,
> the first is "The activator
> org.xtext.example.mydsl.ui.internal.MyDslActivator for bundle
> org.xtext.example.mydsl.ui is invalid") and the input "++ Hi" is not
> parsed correctly. What's wrong in the grammar?

Report message to a moderator

Re: Xtext hangs [message #879804 is a reply to message #879759]

Thu, 31 May 2012 21:48

Oleg Bolshakov

Messages: 36
Registered: August 2010

Member

I was really wrong about the tree and yes, the grammar makes input as though to be infinite.

Sebastian, I didn't increase memory and am not sure how to do it properly, if you mean new Eclipse configurations, so there seem to be default VM parameters for my Eclipse: "-Dosgi.requiredJavaVersion=1.5 -Xms40m -Xmx384m".

Henrik, thank you for you version of grammar, it's not ambiguous as mine.
But what about this one grammar, hanging my Xtext too?

Model:
	elems+=MyRule*;
	
MyRule:
	NameRule
	|'++' MyRule {MyRule.left = current} right = ID
	|'--' MyRule {MyRule.left = current} right = ID
    |'..' MyRule {MyRule.left = current} right = ID
;

NameRule:
	name = ID
;

Here MyRule is not infinite when after ther operator ('++'|'--'|'..') there is an ID (and one else ID for right part) so that it should parse inputs like this:

++ a b ++ c d

[Updated on: Thu, 31 May 2012 21:50]

Report message to a moderator

Re: Xtext hangs [message #879816 is a reply to message #879804]

Thu, 31 May 2012 22:24

Henrik Lindberg

Messages: 2509
Registered: July 2009

Senior Member

I have
-Xmx2000m
for a complex grammar I worked on.

When you say it hangs - how long did you wait? There is a timeout
eventually, I forgot how long it is. I had a problem that the timeout
was to short so I had to bump it. (Since then I have simplified my
grammar considerably).

If you increase memory you may get a decent error message (or a working
grammar).

Do you have backtracking turned on? If you have and grammar is complex
you may hit the timeout and you get a grammar that is not what you expect.

I really don't know why what you wrote becomes very complex to solve.
Have you tried if you combine all of the operators into one?

i.e.
MyRule:
NameRule
| op = ('++'['--'|'..') MyRule {MyRule.left = current} right = ID
;

You really do need to remember the operator anyway, and there is no
difference between the rules. Even if you don't want that (for other
reasons), you could try it and see if that reduced the complexity enough
to get some other feedback (or a working parser).

I suspect you have more things in your grammar than just this. Is it
perhaps just an illustration of the problem? Often problems are
"elsewhere" in a complex grammar, and it would help if you posted the
entire grammar.

anyway... hope the above helps you.

- henrik

On 2012-31-05 23:48, Oleg Bolshakov wrote:
> I was really wrong about the tree and yes, the grammar makes input as
> though to be infinite.
> Sebastian, I didn't increase memory and am not sure how to do it
> properly, if you mean new Eclipse configurations, so there seem to be
> default VM parameters for my Eclipse: "-Dosgi.requiredJavaVersion=1.5
> -Xms40m -Xmx384m".
>
> Henrik, thank you for you version of grammar, it's not ambiguous as mine.
> But what about this one grammar, hanging my Xtext too?
> Model:
> elems+=MyRule*;
>
> MyRule:
> NameRule
> |'++' MyRule {MyRule.left = current} right = ID
> |'--' MyRule {MyRule.left = current} right = ID
> |'..' MyRule {MyRule.left = current} right = ID
> ;
>
> NameRule:
> name = ID
> ;
>
> Here MyRule is not infinite when after ther operator ('++'|'--'|'..')
> there is an ID (and one else ID for right part)

Report message to a moderator

Re: Xtext hangs [message #879954 is a reply to message #879816]

Fri, 01 June 2012 07:41

Oleg Bolshakov

Messages: 36
Registered: August 2010

Member

I've tried to increase amount of available Eclipse memory in eclipse.ini from -Xmx384m to -Xmx1000m, but Eclipse seem to not take more memory when generating workflow. I still have about 230 MB of memory in Eclipse.exe use and about 80 Mb of Javaw.exe (I have 2 GB on my PC and just the half is used actually in this moment by the whole system).

I waited for 10 minutes for the workflow to generate, but in vain.
what I see in console is:

0    [main] INFO  lipse.emf.mwe.utils.StandaloneSetup  - Registering platform uri '<link>'
1122 [main] INFO  ipse.emf.mwe.utils.DirectoryCleaner  - Cleaning <link>
1139 [main] INFO  ipse.emf.mwe.utils.DirectoryCleaner  - Cleaning <link>
1327 [main] INFO  ipse.xtext.generator.LanguageConfig  - generating infrastructure for org.xtext.example.mydsl1.MyDsl with fragments : ImplicitRuntimeFragment, ImplicitUiFragment, GrammarAccessFragment, EcoreGeneratorFragment, SerializerFragment, ResourceFactoryFragment, XtextAntlrGeneratorFragment, JavaValidatorFragment, ImportNamespacesScopingFragment, QualifiedNamesFragment, BuilderIntegrationFragment, GeneratorFragment, FormatterFragment, LabelProviderFragment, OutlineTreeProviderFragment, QuickOutlineFragment, QuickfixProviderFragment, JavaBasedContentAssistFragment, XtextAntlrUiGeneratorFragment, Junit4Fragment, TypesGeneratorFragment, XbaseGeneratorFragment, CodetemplatesGeneratorFragment, RefactorElementNameFragment, CompareFragment
3325 [main] INFO  clipse.emf.mwe.utils.GenModelHelper  - Registered GenModel '<link>' from '<link>/MyDsl.genmodel'

After this part of log for good grammars I used to see the message "generating Java-based EValidator API" which I get about in 2-3 seconds after the last line in above log.

This grammar...

MyRule:
NameRule
| op = ('++'['--'|'..') MyRule {MyRule.left = current} right = ID
;

...gets a syntax error: An unassigned rule call is not allowed, when the 'current' was already created.

This grammar...

MyRule:
NameRule
| ('++'['--'|'..') MyRule {MyRule.left = current} right = ID
;

...adds in about 3 seconds one error message in console: "constraint is INVALID for context MyRule_MyRule_1_2 and type NameRule" and then hangs Eclipse.

This grammar...

MyRule:
NameRule
| ('++'['--') MyRule {MyRule.left = current} right = ID
;

...does not hang Eclipse, adds the previously mentioned error message but after that generates validator and finishes fine in seconds.
The resulting editor seem to work, being able to parse this input "++ aa aa -- bb bb" and showing syntax error for "++ aa".

This grammar...

MyRule:
	NameRule
	| '++' MyRule {MyRule.left = current} right = ID 
	| '--' MyRule {MyRule.left = current} right = ID
;

does everything like the previous one but there is no error in workflow console.

The whole grammar listed in some above message is not used with any extra rules: I created an ampty Xtext project for experimenting with this simple grammars which make Xtext workflow generating process hang (when I said before "hangs Eclipse" I told just about workflow generation process that I can stop in, for example, console view).

P.S. But this Grammar seem to work (correct variant of Henrik's suggestion)!!

MyRule returns MyRule:
	op = ('++' | '--' | '..') left = MyRule? right = ID
;

Yeah..so the point is that if you have some sort of prefix in grammar, when some rule may contain another inside - you should create an AST object as soon as you understood - you have a parent element and then you should just try to calculate the children by assigning its calling rule to a property of just created parent object. So it seems that our grammars were not correct, right?

[Updated on: Fri, 01 June 2012 10:43]

Report message to a moderator

Re: Xtext hangs [message #880053 is a reply to message #879954]

Fri, 01 June 2012 10:36

Henrik Lindberg

Messages: 2509
Registered: July 2009

Senior Member

Maybe something like this is what you want:

MyRule:
NameRule
| op=('++'|'--'|'..') right = ID
({MyRule.left == current}
op=('++'|'--'|'..') right=ID)*
;
NameRule : name = ID ;

Or

Rule : NameRule | MyRule ;
MyRule : op=('++'|'--'|'..') right = ID
({MyRule.left == current}
op=('++'|'--'|'..') right=ID)*
;
NameRule : name = ID ;

Regards
- henrik

On 2012-01-06 9:41, Oleg Bolshakov wrote:
> I've tried to increase amount of available Eclipse memory in eclipse.ini
> from -Xmx384m to -Xmx1000m, but Eclipse seem to not take more memory
> when generating workflow. I still have about 230 MB of memory in
> Eclipse.exe use and about 80 Mb of Javaw.exe (I have 2 GB on my PC and
> just the half is used actually in this moment by the whole system).
>
> I waited for 10 minutes for the workflow to generate, but in vain.
> what I see in console is:
> 0 [main] INFO lipse.emf.mwe.utils.StandaloneSetup - Registering platform
> uri '<link>'
> 1122 [main] INFO ipse.emf.mwe.utils.DirectoryCleaner - Cleaning <link>
> 1139 [main] INFO ipse.emf.mwe.utils.DirectoryCleaner - Cleaning <link>
> 1327 [main] INFO ipse.xtext.generator.LanguageConfig - generating
> infrastructure for org.xtext.example.mydsl1.MyDsl with fragments :
> ImplicitRuntimeFragment, ImplicitUiFragment, GrammarAccessFragment,
> EcoreGeneratorFragment, SerializerFragment, ResourceFactoryFragment,
> XtextAntlrGeneratorFragment, JavaValidatorFragment,
> ImportNamespacesScopingFragment, QualifiedNamesFragment,
> BuilderIntegrationFragment, GeneratorFragment, FormatterFragment,
> LabelProviderFragment, OutlineTreeProviderFragment,
> QuickOutlineFragment, QuickfixProviderFragment,
> JavaBasedContentAssistFragment, XtextAntlrUiGeneratorFragment,
> Junit4Fragment, TypesGeneratorFragment, XbaseGeneratorFragment,
> CodetemplatesGeneratorFragment, RefactorElementNameFragment,
> CompareFragment
> 3325 [main] INFO clipse.emf.mwe.utils.GenModelHelper - Registered
> GenModel '<link>' from '<link>/MyDsl.genmodel'
>
> After this part of log for good grammars I used to see the message
> "generating Java-based EValidator API" which I get about in 2-3 seconds
> after the last line in above log.
>
> This grammar...
> MyRule:
> NameRule
> | op = ('++'['--'|'..') MyRule {MyRule.left = current} right = ID
> ;
> ...gets a syntax error: An unassigned rule call is not allowed, when the
> 'current' was already created.
>
> This grammar...
> MyRule:
> NameRule
> | ('++'['--'|'..') MyRule {MyRule.left = current} right = ID
> ;
> ...adds in about 3 seconds one error message in console: "constraint is
> INVALID for context MyRule_MyRule_1_2 and type NameRule" and then hangs
> Eclipse.
>
> This grammar...
> MyRule:
> NameRule
> | ('++'['--') MyRule {MyRule.left = current} right = ID
> ;
> ...does not hang Eclipse, adds the previously mentioned error message
> but after that generates validator and finishes fine in seconds.
> The resulting editor seem to work, being able to parse this input "++ aa
> aa -- bb bb" and showing syntax error for "++ aa".
>
> This grammar...
> MyRule:
> NameRule
> | '++' MyRule {MyRule.left = current} right = ID | '--' MyRule
> {MyRule.left = current} right = ID
> ;
> does everything like the previous one but there is no error in workflow
> console.

Report message to a moderator

Re: Xtext hangs [message #880057 is a reply to message #880053]

Fri, 01 June 2012 10:46

Oleg Bolshakov

Messages: 36
Registered: August 2010

Member

Henrik, I made a change to the bottom of my last post. I think all our grammars are incorrect because we must create a parent object as soon as we met a prefix ('++' | '--' | '..') and try to find the chilren after that prefix (which can be the same as parent object) and assign it to the field of created one.

So this grammar:

MyRule returns MyRule:
	op = ('++' | '--' | '..') left = MyRule? right = ID 
;

works fine with smth like:

++ -- aa bb

P.S. Sorry, my intention really changed by what I wanted to do first time Smile

. Now I wanted some rule to have some prefix.

[Updated on: Fri, 01 June 2012 10:51]

Report message to a moderator

Re: Xtext hangs [message #880148 is a reply to message #880057]

Fri, 01 June 2012 14:16

Henrik Lindberg

Messages: 2509
Registered: July 2009

Senior Member

On 2012-01-06 12:46, Oleg Bolshakov wrote:
> Henrik, I made a change to the bottom of my last post. I think all our
> grammars are incorrect because we must create a parent object as soon as
> we met a prefix ('++' | '--' | '..') and try to find the chilren after
> that prefix (which can be the same as parent object) and assign it to
> the field of created one.

This:

MyRule : op=('++'|'--'|'..') right = ID
({MyRule.left == current}
op=('++'|'--'|'..') right=ID)*
;

....does just that - when the op is seen, it creates an instance of
MyRule and assigns op, then right. At this point you have parsed

++ abc

Then, if that part is followed by an op, e.g.
++ abc -- def
you get a new MyRule with left set to (++ abc), and its op and right are
set - then it iterates for as long as the MyRule is followed by an op.
For each iteration the tree goes on the left branch and the rightmost
MyRule is at the top of the subtree.

- henrik

Report message to a moderator

Re: Xtext hangs [message #880210 is a reply to message #880148]

Fri, 01 June 2012 16:25

Oleg Bolshakov

Messages: 36
Registered: August 2010

Member

I see now. You're right. How do you think, does it differ:

{MyRule.left == current} op=('++'|'--'|'..')

from

op=('++'|'--'|'..') {MyRule.left == current}

?
I suppose MyRule element to be created just when the operator was found.

P.S. I see it is in the case but I can't delete the message.
in the second case the op feature will be reassigned with a new operator and the newly created MyRule will not contain the corresponding operator.

But does it matter in any other case:

MyRule2:
{Rule1} "keyword" op = ('+' | '-')

and

MyRule3:
"keyword" {Rule1} op = ('+' | '-')

Is there any difference?

[Updated on: Fri, 01 June 2012 16:29]

Report message to a moderator

Re: Xtext hangs [message #880597 is a reply to message #880210]

Sat, 02 June 2012 11:46

Henrik Lindberg

Messages: 2509
Registered: July 2009

Senior Member

On 2012-01-06 18:25, Oleg Bolshakov wrote:
> I see now. You're right. How do you think, does it differ: {MyRule.left
> == current} op=('++'|'--'|'..')

New MyRule created with current as left, new MyRule becomes current, and
op is set in the new.

> from
> op=('++'|'--'|'..') {MyRule.left == current}?
> I suppose MyRule element to be created just when the operator was found.
op is set in the current MyRule, a new MyRule is then created with the
current as left.

The last ? does not make sense (optionally create new instance and
assign left...)

Regards
- henrik

Report message to a moderator

Previous Topic:	resource descriptions in unit tests
Next Topic:	Why this grammar is not LL(*)?

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

]

Current Time: Fri Apr 19 10:22:40 GMT 2024

.:: Contact :: Home ::.

Breadcrumbs

Sign up to our Newsletter