Compilation fails for attribute declaration when incremental + flush + tracing

[René Schöne]

Given the parameters

--flush=full
--incremental=param
--tracing=cache,flush

When defining an (inherited) attribute

Then the compilation fails with the error

org.jastadd.tinytemplate.TemplateExpansionWarning: Template expansion warning: while expanding template 'AttrDecl.emitEquation': while expanding template 'AttrDecl.cacheCheck': while expanding template 'AttrDecl.incHookAttrRead': while expanding template 'AttrDecl.reactToDepChange': unbound variable 'HostClass'

The case is illustrated in https://git-st.inf.tu-dresden.de/jastadd/playground/-/tree/master/nta-token-inc
It uses the grammar

Root ::= WorkingType ErroneousType ;
WorkingType ::= <ID> /A:A/ /<ComputedName:String>/;
A ::= <ID> ;
ErroneousType ::= <ID> ;

and the aspect

aspect Attributes {
  syn A WorkingType.getA() = new A();
  syn String WorkingType.getComputedName() = getID() + "more";
  inh int ErroneousType.ivalue();
  eq Root.getErroneousType().ivalue() = 1;
}

A suggested change is

--- a/src/template/incremental/Notification.tt
+++ b/src/template/incremental/Notification.tt
@@ -235,7 +235,7 @@ $endif
   #resetCache
   #(signature)_handler.notifyDependencies();
   $if (TraceFlush)
-  state().trace().flushIncAttr($HostClass.this, "#signature", "", "");
+  state().trace().flushIncAttr(#hostClassName.this, "#hostClassName.#signatureJavaStyle", "", "");
   $endif
 }
 $endif

‌

[Jesper Öqvist]

I tried adding your test as a standalone test to the regression test suite but I get a different error than the one you describe:

    [junit] Compilation failed when expected to pass:
    [junit] warning: [options] bootstrap class path not set in conjunction with -source 7
    [junit] tmp/incremental/param/nta_token_03/ASTNode.java:523: error: cannot find symbol
    [junit]             res.append("[" + index + (mayHaveRewrite() ? ",r" : "") + "]");

The change from $HostClass to #hostClassName should be good either way.

[Jesper Öqvist]

I tried it with JastAdd2 2.3.4-68-g9a65594c btw.

[René Schöne]

Strange. Can’t reproduce the error you get (running the example from the playground repository) using various build settings.

Maybe one thing I forgot to mention, the attribute in question (ivalue) needs to be marked as cached, either directly or using --cache=all. But I still only get the error I wrote in the description.

[Jesper Öqvist]

What build of JastAdd are you running?

[René Schöne]

Same as you, 2.3.4-68-g9a65594. To test it, I’ve used the playground repository and changed the jastadd2 dependency from "org.jastadd:jastadd:2.3.4" to my locally built jar using files('../jastadd-2/jastadd2.jar').

[Jesper Öqvist]

When I added cache=all I got the same error as you!

[Jesper Öqvist]

After applying your patch my original error remains - the mayHaveRewrites error.

[René Schöne]

Could you commit your changes to some branch in the jastadd-test repo, so I can try to reproduce it? With the playground-repo I’m not able to.

[Jesper Öqvist]

My test was using rewrite=regular, switching to rewrite=cnta fixed the error.

[Jesper Öqvist]

The test harness uses different default options than the gradle plugin.

[Jesper Öqvist]

[incremental] Fix unbound template variable

fixes #332

→ <<cset 1a3072525645>>

[Jesper Öqvist]

The fix caused a test regression. The reason is that #hostClassName is evaluated on an attribute declaration when the host class of the attribute equation needs to be used.

[Jesper Öqvist]

[incremental] Fix unbound template variable

This reverts commit 1a3072525645689f52fce34e02cc3684ec781be2.

fixes #332

→ <<cset 4acf4f072b03>>