FWD » Core Development » Conversion Tools

The ABLUnit/OEUnit @TestSuite annotation lists classes that are to be executed as part of the suite. We will make conversion smarter/easier if we automatically add these to the conversion list.

This root is built in ProgressParser.oea_annotation(), line 10017:

@ [ANNOTATION]:<id_unavailable> @1:1
   TestSuite [SYMBOL]:<id_unavailable> @1:2
   = [ASSIGN]:<id_unavailable> @1:22
      procedures [SYMBOL]:<id_unavailable> @1:12
      "tests.small.start,                       tests.small.start1" [STRING]:<id_unavailable> @1:23

[...]. It is possible to extract the text from the STRING and turn it into a filename which can be searched. Then maybe ScanDriver.addSource() can be used.

Yes, this makes sense. You can use character.parseProgressCharLiteral() to create the Java String instance.

Greg Shah wrote:

[...]. It is possible to extract the text from the STRING and turn it into a filename which can be searched. Then maybe ScanDriver.addSource() can be used.

Yes, this makes sense. You can use character.parseProgressCharLiteral() to create the Java String instance.

The most important part is adding the source to the symbol resolver.

sym.loadClass(files[i], true);

java.lang.RuntimeException: Cannot find class/interface tests/small/start in PROPATH.

So matching the procedure/class to the correct filename is something I need to work on. There's SymbolResolver.initPossibleClasses() which does this, but only for *.cls.

In the following code:

            // find the class file name in the propath
            if (!srchPropathFindFile(found.clsname, found.type, found))
            {
               if (wa.preScanPass >= 1 && !wa.forceLoadFailure)
               {
                  classDict.addSymbol(false, found.clsname, MOCK_CLASS_DEF);
                  wa.preScanRefs.add(name);

                  // this class is not found, but it must not break parsing - show a warning and
                  // 'sink' all references to it
                  String err = reportClassNotFound(found.clsname, false);
                  System.out.printf("WARNING: Lvl%02d parse: %s\n", wa.preScanPass, err);

                  return found.clsname;
               }
               else
               {
                  // we fail only if we are referencing this class from the currently parsed
                  // file (we are not in pre-scan-mode)
                  reportClassNotFound(found.clsname, true);
               }
            }

I'll check where this gets me and expand from there.

Greg, I think we'll need something similar to pre_scan_class in progress.g, but for procedures. What do you think?

Dănuț Filimon wrote:

Greg, I think we'll need something similar to pre_scan_class in progress.g, but for procedures. What do you think?

I would very much like to avoid that. pre_scan_class is only needed because the 4GL compiler uses 2 passes to parse .cls files. In a procedure file (e.g. .p), code cannot reference symbols defined later then itself. So if you define a variable you can only use it after that definition. In a .cls, that is not the case.

Can't you call the same processing directly from the oea_annotation rule? Why does it need to be in pre_scan_class? The files referenced don't have to be loaded before parsing of the current file finishes, so we shouldn't need to do this during pre_scan_class.

Greg Shah wrote:

Dănuț Filimon wrote:

Greg, I think we'll need something similar to pre_scan_class in progress.g, but for procedures. What do you think?

I would very much like to avoid that. pre_scan_class is only needed because the 4GL compiler uses 2 passes to parse .cls files. In a procedure file (e.g. .p), code cannot reference symbols defined later then itself. So if you define a variable you can only use it after that definition. In a .cls, that is not the case.

Can't you call the same processing directly from the oea_annotation rule? Why does it need to be in pre_scan_class? The files referenced don't have to be loaded before parsing of the current file finishes, so we shouldn't need to do this during pre_scan_class.

I need to call ScanDriver.addSource() and this happens when I call SymbolResolver.loadClass(). I can avoid using AstGenerator.preScanClass() entirely and the conversion will work.

The TestSuite can specify both the classes and procedures, so it's better to handle all of them rather than half. The filesystem search is the best option right now because, as mentioned, adding the external procedures when calling loadClass() will cause a lot of confusion.

@TestSuite(classes="").
@TestSuite(classes="   ").
@TestSuite(classes=",").
@TestSuite(classes="tests.TestClass1").
@TestSuite(classes="tests/TestClass2.cls").
@TestSuite(claSSes="tests/TestClass3").
@TestSuite(CLassES="tests/TestClass4.cls,tests/TestClass5.cls").
@TestSuite(CLASSES="tests/TestClass6.cls,tests/TestClass7,tests.TestClass8").
@TestSuite(procedures="").
@TestSuite(procedures="    ").
@TestSuite(procedures=",").
@TestSuite(procEDUres="tests/testproc1.p").
@TestSuite(PRoceduRES="tests/testproc2.p,tests/testproc3.p").
@TestSuite(classes="tests.TestClass9,,tests/TestClass10.cls,tests/TestClass11", procedures="tests/testproc4.p,,tests/testproc5.p").
CLASS TestSuite7851:

END CLASS.

• Only *.p files can be used for procedures,
• In procedures, / must be used and not ..
• In procedures, if the extension is not mentioned, it will use .cls
• Multiple classes/procedures can be specified, but the ones after the first (in order) will always fail to find the first file (regardless if it exists or not).
• , can be used to specify empty entries, it will not affect the list.
• classes/procedures must be used as full words, can't be abbreviated.

+ Marian/Vladimir

I noticed I was not listed as a reviewer, though I have already completed the review.

7851a rev 16299 code review:

1. SymbolResolver.loadProcedure method should be static.
2. progress.g: symbolText is never null, so the null check should be removed:
   

         if (symbolText == null ||
   

3. The code:
   

            boolean isTestSuite = false;
            if (tmp != null)
            {
               isTestSuite = tmp.getText().equalsIgnoreCase("TestSuite");
            }
   

   
   can be rewritten as:
   

            boolean isTestSuite = tmp != null && tmp.getText().equalsIgnoreCase("TestSuite");
   

Also, from your code I gathered the classes in the @TestSuite annotation can be referenced by class file names instead of class names. This is not mentioned in OE ABLUnit documentation, but it works in OE.
Currently this is not implemented in conversion, but probably should. I am going to create another task for this.

Code Review Task Branch 7851a REvisions 16299 and 16300

The changes are good.

Greg Shah wrote:

+ Marian/Vladimir

Greg, all I can share about this is that we were only running `ABLUnit` in FWD from inside Eclipse so we always excluded the `test suites` from conversion just to avoid having all tests ran twice. Not sure if Vladimir had changed anything in that area but at the time the engine was running all tests in the package so if there were a test suite all tests were executed twice, hence we didn't bother with the test suite nor had the need to automatically add the tests in it to the conversion list but this makes a lot of sense if the engine allows specifying a test suite for test execution.

What we had done on the subject was to automatically add all base classes, statically ran procedures and database table triggers to the conversion list - everything that we've considered as 'dependency' in order to keep the conversion lists cleaner. Any dynamic run - using `value` or `call` interface - can't be resolved as it can actually be anything if variables are used :(

Regression tests passed, I am rebasing to check for conflicts.

Rebased 7851a to latest trunk/16312, the branch is now at revision 16315.

7851a can be merged.

Trunk is currently frozen for a delivery.

Alex: Please schedule this in the merge queue when trunk thaws.