Runtime Error Messages

Runtime stack traces

Stack trace errors commonly look like this:

(DN.307): While evaling syn 'silver:analysis:typechecking:core:upSubst' in 'silver:definition:core:presentAppExpr' (1b842264, Command.sv:98:30)
(DN.307): While evaling syn 'silver:analysis:typechecking:core:upSubst' in 'silver:analysis:typechecking:core:check' (53df4528)
(silver.definition.type.PunifyCheck in PunifyCheck.java:116): Error while evaluating function silver:definition:type:unifyCheck

There are a few things to note about how to read this output:

  • They begin with something in parenthesis. For things in the Silver Runtime, this is shortened to an abbreviation (like DN.307). For things coming from generated code, there is a full package and class name, followed by a file name and line number (like silver.definition.type.PunifyCheck in PunifyCheck.java:116).

  • Each production is identified by the System.identityHashCode value of its decorated node, and (if present at all) the location annotation attached to that node. The identity allows spotting the same node in the stack trace, and the location is occasionally helpful when tracking down bugs with your program: often it’s easier to diagnose a problem when you know what bit of syntax lead to it.

  • Begin reading stack traces at the bottom: that’s often where the only relevant bit is.

  • Silver stack traces do NOT show thunk evaluation, which means they can occasionally be misleading, as thunks means evaluation can “jump” elsewhere. The underlying (large!) Java stack traces will be printed instead if you set the SILVER_TRACE environment variable to 1.

Stack trace messages

Inherited attribute _a_ not provided to _p1_ by _p2_

Unfortunately this error provides little clue about how the production p1 appeared in its parent node (p2). This is annoying, because the problem likely needs to be fixed by changing p2, somehow.

It could be a child, a local, or an “anonymous” decorate site (using the decorate foo with {...} syntax). Or, one may have written x.foo on an undecorated value x, which implicitly creates a decorated node (with no inherited attributes!) on which to evaluated foo.

If the cause is not obvious, begin by tracking down which nonterminal the production p2 belongs to, and then track down all aspects of the production p1. Look for anywhere a value of that appropriate type might be accessed.

It may also be helpful to run the MWDA on your grammar, which will warn statically about places where inherited equations may be missing. This should make it much easier to find the cause of the problem… if you’re not swamped by many MWDA problems. See how to run the MWDA.

While evaling inh _a_ for forward in _p_ / While evaling syn via forward in _p_

Neither of these tracing message indicate which attribute is being evaluated, as that’s a bit redundant. Look below them until forwarding ends and you should see something like:

(DN.313): While evaling syn via forward in 'silver:definition:core:application' (5a659a05, Command.sv:98:30)
(DN.313): While evaling syn via forward in 'silver:definition:core:functionApplication' (217bd515, Command.sv:98:30)
(DN.307): While evaling syn 'silver:analysis:typechecking:core:upSubst' in 'silver:definition:core:functionInvocation' (19a9b1e9, Command.sv:98:30)

Here, we’re accessing upSubst from an application which forwarded to functionApplication which forwarded to functionInvocation, where there was an equation answering.