[vhdl-200x] conditional compilation response

Hi,

I've had a chance to review the 26 emails on my "conditional
compilation" proposal. I feel as if Evan represented the issue as I
would in his responses. I thank him for that and I'm sorry I don't have
the bandwidth to respond as frequently or eloquently. I would like to
get to some of his early feedback and questions for me, but first things
first.

The objective here is to address problems that conditional compilation
mechanisms solve. There is code users create that will not compile for
a variety of reasons. It may be, as some have noted, problems with
compilers for simulation, synthesis, static analysis, etc. not accepting
some language constructs. This can be bugs, missing or non-standard
features. An extremely common reason is incompatibilities with language
versions. We have this problem with VHDL 2008 versions pre-2008
versions. It is driven by standard package reorganization, basic type
system changes, and lack of support for features. Let me repeat, code
that will not compile consistently. It is a barrier for organizations
that have complex tools flows and need to solve problems that this
mechanism solves.

Conditional compilation is a construction technique for code that
supports other purposes for users. It can be experimentation, dealing
with mixed language and mixed signal composition issues, too. It has
been an often requested enhancement for VHDL as people have pointed out.
HDL is handled very similar to other high level languages. Software
configuration management and version control system are applied with
similar variations process maturity. Conditional compilation mechanisms
can be a very versatile feature. There is a reason it has been requested
for VHDL repeatedly over the last 30 years. I come at this for VHDL very
conservatively.

I am EDA software developer who builds HDL based design tools. I
understand VHDL, compiler design, and a fair amount about high level
language design. I have world class experts at my disposal in these
areas. That said, I only know 2 basic mechanisms used for this purpose.
Simple pragmas and a set of preprocessing directives. They have the
essential property that they are self consistent and do not require the
affected code be analyzed. They also can be applied with fine
granularity to any subset of language they support, i.e., at the token
level. This is a pretty robust thing for dealing with code that will
not compile consistently.

There is no standard way to do this for VHDL. EDA Vendors have created
proprietary solutions to suit their tool purpose and in narrow cases,
the user community has forced defacto support. These are in the form
of simple pragmas and they lack richness in control structure of a
modern set of preprocessing directives. That and the lack of portability
is a long standing concern. It could happen again for vhdl 2008.

This is not a problem to be solved by a first class language feature.
The idea that extending generate is appropriate has been debated here.
The purpose of the generate feature is to support parameterizing design
structure. It uses conditional and interative forms to express such a
design. At analysis, it is all first class language that must fully
conform to grammar and semantic rules. It is subject to visibility and
overloadable resolution rules of the language. And when it is analyzed
,legal and correct design units exist. They are not finalized until
elaboration when the static conditions that govern the generate are
applied.

This is incompatible with the idea of code that cannot compile
consistently. Whatever problems you solve, generate won't be a robust
solution for this one. Moreover, it will never have the granularity to
deal with all of the problems outlined here. I don't wish to insult the
idea or brainstorming about the issue, but I am offering a strong
critique that no first class language extension can properly solve the
problem.

I do feel it is far better to address this with a preprocessing language
than a narrow and specific pragma. If this group thinks differently,
including that is even unworthy to solve this in a standard way and
rejects the requirement, so be it. I would not mind talking about
syntax and feature considerations in a preprocessing language. Evan
had a lot of questions and opinion about that in his initial reply. I
saw that as support for solving the problem and basic approach, and
starting with the devil is the details. You can badly design anything,
including a preprocessing language, but I presume we will do a good job.
There is a lot of precedent to learn from. Let us see if we even get
there.

I will say that preprocessing itself is an implementation decision. It
can be done in the context of the lexical analysis phase of the analyzer
and often is. The real requirement is that it is not encumbered with
first class language features and is compatible but orthogonal to the
language syntax. Standardizing it for vhdl has real value for portable
code composition. I only expressed it in my proposal as a preprocessor
with a subset of C preprocessing features to make the high level ideas
clear.

Regards, John

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.