Operators

[Brian Tiffin]

Finally taking a more serious stab at the Operators chapter of the UP docs.

I may have read the precedence table in src/unicon/unigram.y wrong

expr8   : expr9 ;
        | postfixthreadop ;
        | expr9 CARET expr8 { $$ := node("Bcaret", $1,$2,$3);} ;

postfixthreadop:
          expr9 SND { $$ := node("Bsnd", $1,$2,EmptyNode);} ;
        | expr9 SNDBK { $$ := node("Bsndbk", $1,$2,EmptyNode);} ;
        | expr9 RCV { $$ := node("Brcv", $1,$2,EmptyNode);} ;
        | expr9 RCVBK { $$ := node("Brcvbk", $1,$2,EmptyNode);} ;

expr9   : expr10 ;
        | expr9 BACKSLASH expr10 { $$ := node("limit", $1,$2,$3);} ;
        | expr9 AT expr10 { $$ := node("at", $1,$2,$3);} ;
        | expr9 SND expr10 { $$ := node("Bsnd", $1,$2,$3);} ;
        | expr9 SNDBK expr10 { $$ := node("Bsndbk", $1,$2,$3);} ;
        | expr9 RCV expr10 { $$ := node("Brcv", $1,$2,$3);} ;
        | expr9 RCVBK expr10 { $$ := node("Brcvbk", $1,$2,$3);} ;
        | expr9 BANG expr10 { $$ := node("apply", $1,$2,$3);};

unigram.y

I take that as placing the postfix >@ send operator (and friends) at a lower precedence than the infix form. But now I'm thinking they are all at "level 9", (or are they wedged in between at 8b)?

See the precedence chart (based on the ProIcon docs and modified) at

http://btiffin.users.sourceforge.net/up/operators.html

Is exponentiation grouped with e1 >@ or should postifx >@ be grouped with infix e1 >@ e2 ?

I think I have this wrong, and could use some expert advice.

Cheers,
Brian

[Bruce Rennie]

Good evening Brian,

The @>, @>>, <@ and <<@ have problematic precedence as they are unary
prefix, unary postfix, binary and operandless. The actual design is

[x] op [y]

with both x and y optional. Hence, giving rise to unary prefix, unary
postfix, binary and operandless. Based on previous discussion (thanks
Jafar), they should be considered binary with the missing operand
defaulting to a specific value for each case. The precedence of each
operand is spread over a number of levels, more than just between expr8
and expr9 (you also need to include expr11 as well).

I, personally, think that the grammar description needs more thought and
clarification. But at this point, I am not sure what should be proposed.
I am having a look at various renderings to see what happens, but I am
not far into this process.

One proposal so far is

exprx :

       expr9  SND  {  $$  :=  node("Bsnd",  $1,$2,EmptyNode);}  ;
     |  expr9  SNDBK  {  $$  :=  node("Bsndbk",  $1,$2,EmptyNode);}  ;
     |  expr9  RCV  {  $$  :=  node("Brcv",  $1,$2,EmptyNode);};
     |  expr9  RCVBK  {  $$  :=  node("Brcvbk",  $1,$2,EmptyNode);}  ;
     |expr9  SND  expr10{  $$  :=  node("Bsnd",  EmptyNode,$2,$3);}  ;
     |  expr9  SNDBK  expr10 {  $$  :=  node(Bsndbk,EmptyNode,$2,$3);}  ;
     |  expr9  RCV  expr10{  $$  :=  node("Brcv",  EmptyNode,$2,$3);}  ;
     |  expr9  RCVBK  expr10{  $$  :=  node("Brcvbk",  Emptynode,$2,$3);}  ;
     | SND{  $$  :=  node("Bsnd",  EmptyNode,$1,EmptyNode);}  ;
     |  SNDBK  {  $$  :=  node("Bsndbk",  EmptyNode,$1,EmptyNode);}  ;
     |  RCV  {  $$  :=  node("Brcv",  EmptyNode,$1,EmptyNode);};
     |  RCVBK  {  $$  :=  node("Brcvbk",  EmptyNode,$1,EmptyNode);}  ;

But I don't know what ramification this would have.

Jafar has clarified some questions I have had. I still don't have
enough clarity as to what is a cleaner set of productions.

regards

Bruce Rennie

[Brian Tiffin]

Thanks, Bruce. Clarifies a little bit. I'm going to leave the chart as-is for the moment, perhaps add the exceptions in prose following the table.

Cheers,
Brian

[Jafar]

You are correct Bruce. I view the communication operators as binary operators, missing right operands defaults to the current thread , while missing left operand defaults to null. If I remember our initial design, the intention was all forms have the same priority with them being resolved left to right.

Cheers,
Jafar

[Jafar]

Of course, as you and I discussed before Bruce, default left operand make sense with @> and @>> where you just send &null. with <<@ it the left operand is a timeout of how long you are willing to block, but there is no intuitive clear meaning for <@. I have experiementd with a few but was not satisfied. If you can think if some good use for it please let me knwo.

--Jafar

[Bruce Rennie]

I am thinking of making a change to the grammar as I sort of outlined
earlier and see what this does with the current system. But I won't be
ready for this until after I have finished the things currently on my
plate, including a new lexer sub-system for the regexp facility.

regards

Bruce Rennie

[Charles Evans]

Great resource, Brian, thanks!

[Charles Evans]

Could you put a link to your string scanning page in the "?" operator section? Thanks.

[Brian Tiffin]

Umm, two years later, ... noticed.

Should be fixed in 0.6.148. Which might contain a few loose edges at the moment, a reindexing work is in progress.

https://unicon.sourceforge.io/up/operators.html#string-scan

Sorry, Charles, and thanks for the hint.

Have good