Hi Akim;
After a while I decided, why not. I might just as well answer you.
Thanks for your response. This will be a summary and not a
point-to-point answer. It is clear that Bison does not want to, and
will not, proceed forwards, and so, take this as me just grumbling in
my cave.
I do think the effort is moderately difficult. I do not believe it is as
hard as you have indicated. The languages you generate Bison for is
modest, and the semantics are similar (a caveat here since I don't know
D). If you don't want to develop you're own parser, then you can always
fiddle with the gcc (https://gcc.gnu.org/) front end to get a desired
result.
"... fragile magic ..." Ahem. No sense going into hyperbole. This one is
out-of-line and unnecessary. The idea proposed is to construct a
dependency graph, usage graph, of variables with respect to our bad boy,
and to do code reorganization based on the results of the graph. Now the
first book on optimization I read was written in about 1969 (Cock and
Schwartz) and I believe had a dependency graph. Some compilers allow
output of a dependency graph, and certainly optimization requires it.
Unless an implementation is poor and ill-defined in some sense, it
shouldn't be magic and it should be robust in the developers sure hand.
As to "confusing", if you say so. But, if it works then it is a "black
box" to the users. If it doesn't work then the user has done something
wrong, and this should not be part of it. Let's look at an optimizing
compiler for a moment. Optimization reorganizes code on a regular basis.
Someone looking at the code generated will certainly be confused by what
is seen, and this is no more confusing than the proposed method.
The substantive points brought out are well taken, in particular, that
Bison allows output to heretofore unknown languages by use of a
skeleton. Good point in that automatic generation of a dependency graph
depends on language semantics and an unknown language has no Bison known
semantics.
Lest we get carried away, this is your product and you are free to do as
you wish. My history with your product, and its predecessor, Yacc, is
that I am impressed. It is very, very good.
Please ignore what you want.
caiou
art
On 9/4/2022 1:23 AM, Akim Demaille wrote:
Hi,
Le 20 juin 2022 à 22:02, slipbits <h...@slipbits.com> a écrit :
Hi Kaz;
Sorry about taking so long to respond.
Compared to my speed, you are almost relativistic.
By tag, do you want to remove the %code <qualifier> feature?
From where?
It looks like it is possible to remove all <qualifier>s from all %code blocks.
This is because the Bison generated statements are completely in control 0f
Bison and since Bison has knowledge of the working environment, Bison can
elect to move the contents of the declarations in %code blocks with the aid
of a dependency graph.
IIUC, you are suggesting that Bison should guess where to place the various
code blocks, depending on their content instead of their name/tag.
There's a number of reasons that go against this:
- first and foremost, code blocks are black boxes to Bison. It does not parse
them, and never will. That would require whole (compiler) frontends for all
the languages we support, which will never happen.
- the "dependencies" might be hidden. For instance in C, a #include might be a
provider of something another would be consumer of. Sure, if we have a complete C
frontend, the problem is solved :)
- we actually invented named blocks to depart from a model where Bison
"guessed" where to issue the user's code based on some heuristics. It was
confusing, resulted in many users puzzled by drastic changes in the generated parser
because of apparently minor reordering in the source file.
- being picky is definitely not very novice-friendly, but predictability is far
more important. Dumb but clear rules are often superior to fragile magic.
- many of these code blocks are specific to the target language, and baking
into bison-the-executable the resolution of these dependencies would ruin the
fact that today people can have their own skeletons, possibly for languages we
are unaware of.
I'm sure that are other issues.
What I am saying is that it looks like it is possible.
I don't believe it is. I can perfectly understand the desire to have a thiner
documentation, and more automation, but in this precise case, I'm fairly
confident the implementation cost would be enormous, for a marginal win.
Cheers!