Verilog and SV seem to have consciously avoided specifying
preprocessing steps in any detail.
Compare our text with the C and C++ standard translation phase
specification is like comparing
fog to sunshine. Given that, any reading for "original intent" is
delusional at best, but here goes...
Both section 25.5.1:
The macro text can be any arbitrary text
specified on the same line as the text macro name. If more than one
line is necessary to specify the text, the newline shall be preceded by
a backslash ( \ ). The first newline not preceded by a backslash shall
end the macro text. The newline preceded by a backslash shall be
replaced in the expanded macro with a newline (but without the
preceding backslash character).
... and section 5.4:
A one-line comment shall start with the two
characters // and end with a newline.
... presume to own the input stream until the next newline character.
Unlike C/C++, backslash line continuation is not described as taking
the highest precedence. In fact it
is described exactly twice -in 25.5.1 (above) and in 5.9 quoted below
A string literal shall be contained in a single
line unless the new line is immediately preceded by a \ (backslash). In
this case, the backslash and the new line are ignored.
[Note to editor: you don't really mean that the new line is
ignored! Only the newline character; please compound those
words]
In these two cases \ has selectively appeared as a special syntax for
producing two special kinds of BNF nonterminals. As a corollary, we
expect that trying to break an
expression across multiple lines
with a backslash continuation is not a transparent noop - it will
usually produce a syntax error for the unknown \ operator. Similarly,
the lack of a line continuation rule for one-line comments suggests
that the comment text absorbs the backslash character.
Having thus violated one of the earliest Unix conventions, Verilog has
been free to be its own unique kind of crazy ever since.
Implementations seem to be free to give precedence to either one-line
comment or macro text in these cases. I'm strongly in favor of letting
the market decide which precedence is best.
- If // comes first, we might wonder whether the newline delimits both
the comment and the macro text. I think 22.5.1 tried to resolve this
by saying "If a one-line comment (that is, a comment specified with the
characters //) is included in the text, then the comment shall not
become part of the substituted text.". This was necessary for lines
like `define asciiNL 8'h0A // Newline character
- If the macro text sees the line continuation first, the newline
that it leaves in its place gives us a delimiter for the one-line
comment, and the above sentence seems superfluous.
In neither case would the macro pick up any extraordinary power to eat
the remainder of a line on which it was invoked. The LRM prohibition
on one-line comment text substitution doesn't address comment text
which might be separately recognized as a pragma (a non-standard
extension). Those can continue to be substituted with the macro text,
so long as they don't accidentally eat the ends of macro invocation
lines.
I think either of these readings are consistent with the current text
in both letter and spirit. Reading 1 terminates the macro text
immediately before the first one-line comment; Dan's example fails
because a $display() call cannot appear in $unit. Reading 2 allows
Dan's example to parse as legal Verilog, and to cause $display() calls
to happen.
A C/C++-like reading (early folding of line continuation) also makes
Dan's example legal, but the $display() then
becomes part of the comment text.
Greg
Disclaimer: I offer no voting position here - just checking that our
diverging systems are within the letter of the law.
Arturo Salz wrote:
Question
is how should compiler treat comments in precompiler macros - i've
browse
LRM and I haven't found any answer. LRM states only :
If a one-line comment (that is, a
comment
specified with the characters //) is included in the text, then the
comment
shall not become part of the substituted text. The macro text can be
blank, in
which case the text macro is defined to be empty and no text is
substituted
when the macro is used.
But
what
for below case - "//" comment out the end of line
'\" needed to split lines of define - so imho comments finishes the
macros
and below code should fail
CODE:
`define
DAN \
// test \
// test \
$display("ok");
`define DAN1 //ref \
\
$display("ok1");
module top;
initial begin
`DAN
`DAN1
end
endmodule
--
This message has been scanned for viruses and
dangerous content by MailScanner,
and is
believed to be clean.
--
This message has been scanned for viruses and
dangerous content by
MailScanner,
and is
believed to be clean.
--
This message has been scanned for viruses and
dangerous content by
MailScanner, and is
believed to be clean.
Received on Thu Jun 25 15:04:22 2009