Detecting inconsistent selections on Multi Value Enum Attribute via RegExp

llandale - Tue Feb 07 12:16:06 EST 2012

[1] From a theoreticaly viewpoint its likely you are mixing two different pieces of information into one attribute.

[2] I dislike regular expressions; partly because I'm terrible at them and partly because they are so hard for me to read and partly because they are so hard for the next dude to read.

[3] This surely is easier to code with logic if you have more than one error condition.

bool b1 = isMember(o.NameAttr, "v1")
bool b2 = isMember(o.NameAttr, "v2")
bool b3 = isMember(o.NameAttr, "v3")
bool b4 = isMember(o.NameAttr, "v4")
bool bx = true
if ( b1 and  b2  and  b3 and  bx) then "v2 disallowed with v1 and v2"
if (!b1 and  b2  and  bx and !b4) then "v2 disallowed when v1 and v4 are off"
... more conditions ...

-Louie

Helko - Wed Feb 08 07:09:39 EST 2012
Hi Louie,
yes, it is a mixing of two information into one attrbute, but it is a sort of additional constraint for the selection value range.
For me thats in general a common practice. It's not a simple but a complex attribute.

I aggree with you. I also did not like reading regular expressions.

Nevertheless, I need to use it, because the constraints I'm talkiong about here are mainly facing a predefined subset of enum values in this enumeration. At DOORS runtime the DXL program, which is also checking the constraints for each object, adds further enum values in the course of time.

Thus, statical hard coded checks as you suggested with your if statements are not sufficient in my use case.

An example: I have to check, if only "v1" and "v3" are selected but not "v2" or some of the runtime-defined enum values I have mentioned above.
Cheers, Helko

Mathias Mamsch - Wed Feb 08 07:56:01 EST 2012

As Louie I also do not see the need to use regexp on that. But if you still do. Well in theory it should be easy. Note that in DOORS attribute values will have no duplicates, appear in order and separated by \n. Imagine you have three values v1,v2,v3. Then you have 2^3 = 8 possibilities for selections:
 

nothing
v1
v2
v3
v1,v2
v1,v3
v2,v3
v1,v2,v3

Now you can make a regexp:

"(^$)|(^v1$)|(^v2$)|(^v3$)|(^v1\nv2$)|(^v1\nv3$)|(^v2\nv3$)|(^v1\nv2\nv3$)"

which contains all possible values. If you now select 5 of them to be valid, then you remove the other 3 from the regexp and you have a regexp matching your valid patterns (you could also keep the 3 invalid ones) to have the regexp matching invalid combinations. I think you get the idea. It gets problematic if you have 10 enumeration values (1024 possibilities), but only care about 3 of them. In this case you should use Louies approach to preselect the ones you are interested in.

Buffer buf = create() 
 
int count = 0
string val; for val in skEnumValuesIAmInterestedIn do {
   if (isMember(o.NameAttr, val)) {
      if (count++ > 0) buf += "\n"
      buf += val
   }
}

You could then apply your magic regexp on the result again. However might be more efficient instead of using strings to replace your enum values by numbers. Maybe that helps, regards, Mathias

Mathias Mamsch, IT-QBase GmbH, Consultant for Requirement Engineering and D00RS

Helko - Wed Feb 08 08:04:41 EST 2012
Thanks!

Comeing back to the Why:
I cannot check with isMember on enum values which are not present at the beginning or will be defined later on runtime with no limitation on the naming.
Yes, I can check with isMember my predefined subset of enum values.

Thats the reason, why I'm using regular expressions. See my example above. Also the selection status of some runtime defined enum value is relevant for the constraint checking.

SystemAdmin - Wed Feb 08 08:36:50 EST 2012
… and why are you not able to add values to the skip list during run time?
At some point of the script you could ask the current attribute's definition for its defined values (or whatever is your source of information about values to regard) and add them to your skip list, you can define clusters of values which trigger "is allowed or not", use isMember to concat this information and then use some regexp …
but coming back to your original question and the example you gave:
Would the following be a solution for you (yes, this disregards that a value v3a might exist)?

code
Regexp reV1AndV3 = regexp "v1.*\nv3"
Regexp reV1AndV3AndNothingElse = regexp "^v1\nv3$"
if (reV1AndV3 and ! reV1AndV3AndNothingElse) {complain}
code

Helko - Wed Feb 08 08:04:41 EST 2012
Thanks!

Comeing back to the Why:
I cannot check with isMember on enum values which are not present at the beginning or will be defined later on runtime with no limitation on the naming.
Yes, I can check with isMember my predefined subset of enum values.

Thats the reason, why I'm using regular expressions. See my example above. Also the selection status of some runtime defined enum value is relevant for the constraint checking.

Mathias Mamsch - Wed Feb 08 08:45:59 EST 2012

If you do not want to use isMember you can replace it by a findPlainText in a string. I still do not get what you mean by "defined later on runtime with no limitation on the naming". Are you not reading the values from an attribute or doing any postprocessing to the attribute values? However you can always replace the isMember by a findPlainText. The gist is, that you need to extract a subset of your values (i.e. remove all values you do not care for, or put them to separate strings) to use a regexp for checking.

Example for checking for the existence of one value in a string (use that in the filter code of my earlier post):

// if (isMember (o.strAttributeName, val)) { 
    //    ...
    // }
 
    // equivalent code: 
    string check = "\n" (o.strAttributeName) "\n"
    int d1, d2
    if (findPlainText(check, "\n" val "\n", d1,d2, false)) {
        ... 
    }

You might also try to explain a little more, what you are trying to accomplish. Maybe that helps, regards, Mathias

Mathias Mamsch, IT-QBase GmbH, Consultant for Requirement Engineering and D00RS

Helko - Wed Feb 08 09:14:40 EST 2012

OK, again I try to explain by using other words :) :

The module has an enumeration type. An attribute is facing this type.

Before using it daily a set of predefined enum values already exists. These enum values are: "v1", "v2" and "v3".

The end user runs a DXL program in context of the module. At runtime, the user adds further enum values. The value must be unique.

As I said at the beginning, there are additionaly constraints limiting the possibility of selection for a module object.

E.g. It is allowed to have "v1" selected. In that case, "v2" might optionaly be selected, but it is not necessary. It is not allowed to have other enum values selected when "v1" and possibly "v2" is selected. Thus "v3" is forbidden in that case and also other enum values which will be added to the enum type by DXL programm, I've mentioned at the beginning.

There are further so called miss use cases for selections regarding "v1", "v2" and "v3" in special but also regarding the enum values which has been added to the enum type later on.

I try to identify such inconsistencies by checking them per object in the module with certain regular expressions.

E.g.

Regexp missUseCase1= regexp "(v1\n)(v2\n|[^v2].+)";

should be my regular expression for the miss use case example above.

The problem:

[^v2]

excepts the character v and 2, but not the string v2.

Why I'm not using imperative code with a lot of if isMember-checks? For this I have to gather all enum values currently defined in the enum type. Than I have to check using isMember per object if v1 and possibly v2 hase been selected but not (v3 OR some of the user defined enum values).

I'm thinking that the cost and complexity behind such code is more ugly than using regular expressions? That's why I'm trying to find a solution using regular expressions.

But, if it is not possible to define an exception for whole words instead of single characters, I have to realize the if-isMember-checking-solution.

Cheers, Helko

Helko - Wed Feb 08 07:09:39 EST 2012
Hi Louie,
yes, it is a mixing of two information into one attrbute, but it is a sort of additional constraint for the selection value range.
For me thats in general a common practice. It's not a simple but a complex attribute.

I aggree with you. I also did not like reading regular expressions.

Nevertheless, I need to use it, because the constraints I'm talkiong about here are mainly facing a predefined subset of enum values in this enumeration. At DOORS runtime the DXL program, which is also checking the constraints for each object, adds further enum values in the course of time.

Thus, statical hard coded checks as you suggested with your if statements are not sufficient in my use case.

An example: I have to check, if only "v1" and "v3" are selected but not "v2" or some of the runtime-defined enum values I have mentioned above.
Cheers, Helko

Helko - Wed Feb 08 09:14:40 EST 2012

OK, again I try to explain by using other words :) :

The module has an enumeration type. An attribute is facing this type.

Before using it daily a set of predefined enum values already exists. These enum values are: "v1", "v2" and "v3".

The end user runs a DXL program in context of the module. At runtime, the user adds further enum values. The value must be unique.

As I said at the beginning, there are additionaly constraints limiting the possibility of selection for a module object.

E.g. It is allowed to have "v1" selected. In that case, "v2" might optionaly be selected, but it is not necessary. It is not allowed to have other enum values selected when "v1" and possibly "v2" is selected. Thus "v3" is forbidden in that case and also other enum values which will be added to the enum type by DXL programm, I've mentioned at the beginning.

There are further so called miss use cases for selections regarding "v1", "v2" and "v3" in special but also regarding the enum values which has been added to the enum type later on.

I try to identify such inconsistencies by checking them per object in the module with certain regular expressions.

E.g.

Regexp missUseCase1= regexp "(v1\n)(v2\n|[^v2].+)";

should be my regular expression for the miss use case example above.

The problem:

[^v2]

excepts the character v and 2, but not the string v2.

Why I'm not using imperative code with a lot of if isMember-checks? For this I have to gather all enum values currently defined in the enum type. Than I have to check using isMember per object if v1 and possibly v2 hase been selected but not (v3 OR some of the user defined enum values).

I'm thinking that the cost and complexity behind such code is more ugly than using regular expressions? That's why I'm trying to find a solution using regular expressions.

But, if it is not possible to define an exception for whole words instead of single characters, I have to realize the if-isMember-checking-solution.

Cheers, Helko

llandale - Wed Feb 08 16:10:52 EST 2012

No real difference between hard-coding if-elseif logic and hard coding regular expressions. But I think you are saying there is some data-driven constraints for this project that somehow say:

• For attr "X", if "v1" and "v2" are selected then its invalid if "v2" is also selected.

Maybe that's in a table that looks like this:

Attribute    Danger Combo       Invalid Values
X               v1, v3          v2
X               v2, v3, v4      v1

If the Danger Combo exists then it's invalid if any of the Invalid Values may also exist.

for every row in the table
{  NameAttr = Attibute from table
   IsDanger = true
   for each Danger Combo Value
   {  if (!isMember(obj.NameAttr, DangerValue)) IsDanger = false
   }
   if (IsDanger)
   {  IsInvalid = false
      for each Invalid Values
      {  if (!isMember(obj.NameAttr, InvalidValue)) IsInvalid = true
      }
   }
   if (IsInvalid) whatever
}

-Louie

Helko - Thu Feb 09 05:39:02 EST 2012
Dear Matthias, Dear Louie, Others

first: thanks for your answers.

I've understood what you are suggesting.
But the problem is still challenging, because you are argumenting in your examples only based on a predefined set of enum value. But my consistency rules are also partly depending on the selection resp. no selection of enum values the user has made later on (no dependency on certain enum values, but on "if the user has made any selections on user defined enum values").

Mainly the consistency rules are focusing the 3 predefined enum values which are each time on the first 3 positions of the enum value list and each time has the same ordering.

I have no chance to introduce further attributes etc.. That is out of scope of my decision range in the project I'm working for.

Due to the missing power of features when using regular expression in DXL, I've realized my functionality with a lot of if-findPlainText-checks as Matthias suggested yesterday.

Many thanks + cheers, Helko