Improvements to .tst LSP

[aytey]

Collection of miscellaneous areas for improvement:

• (could be easy) We mark withinTest as true if we see TEST.NEW/TEST.REPLACE/TEST.ADD, but the diagnostic when we see TEST.END (without any of the previous three) only lists TEST.NEW and TEST.REPLACE (i.e., it should also list TEST.ADD)
• When creating TEST.VALUE lines, after the first single colon (i.e., after TEST.VALUE:), if the item is a structural (i.e., it contains fields), then we start completing on : incorrectly (it should be completing on ., which creates "nonsense" lines (e.g., TEST.VALUE:uut_prototype_stubs.stub.return:f:f:<<MAX>>) -- for this one, we need to make sure we're at a "scalar" (or similar) before we start accepting : (Andrew thinks this might be tricky)
• When creating TEST.VALUE lines, after the second single colon (first: after TEST.VALUE; second: after unit.sub.param), the extension keeps giving suggestions for additional colons (this allows you to create lines such as TEST.VALUE:unit.subprogram.param:<<MAX>>:<<MAX>>:<<MAX>>) -- for this one, if we see a third single colon, we need to not provide any suggestions
• (could be easy) Auto-complete on void* should list all of the user globals + NULL
• (could be easy) TEST.THISDOESNTEXIST before TEST.NEW gives an error diagnostic; doing it after, the LSP happily ignores totally incorrect TEST.<whatever> commands
• (could be easy) Should we give errors on TEST.REQUIREMENT_KEY if there's no RGW configured?
• (could be easy) For a C++ environment TEST.VALUE.unit.<<GLOBAL>>. lists the "C" instances (e.g., C_976400856_9), these shouldn't be shown
• For pointers:
  • Auto-complete immediately lists p[0], when it should really show p
  • There should be suggestions for <pointer><colon>, which should should list NULL or a malloc size
    • For C++, it should list possible class instances as well (see: Improvements to .tst LSP #108 (comment))
  • Selecting a malloc size should (maybe) create all of the elements in the list
  • If you try to access a cell that isn't in the range of the malloc, we should give an error diagnostic
• We don't give hover information for pointers (either as params or fields)

[aytey]

This example is "abusive" but:

class Meep {
public:
  virtual void meep() = 0;
};

namespace Moo {
class Meep : public ::Meep {
public:
  virtual void meep();
};

namespace Quack {
class Meep : public ::Meep {
public:
  virtual void meep();
};
namespace Oink {
class Meep : public ::Meep {
public:
  virtual void meep();
};
namespace Honk {
class Meep : public ::Meep {
public:
  virtual void meep();
};
class Baa {
public:
  class Meep : public ::Meep {
  public:
    virtual void meep();
  };
};
} // namespace Honk
} // namespace Oink
} // namespace Quack
} // namespace Moo

void foo(Meep &m) { m.meep(); }

We should get:

Right now, we just get the pointer case ("Auto-complete immediately lists p[0], when it should really show p").