This site is a static rendering of the Trac instance that was used by R7RS-WG1 for its work on R7RS-small (PDF), which was ratified in 2013. For more information, see Home. For a version of this page that may be more recent, see WG1Ballot7Results in WG2's repo for R7RS-large.


2012-10-01 21:00:47
1adding seventh ballot resultshistory

Notes about Results

See WG1BallotExplanation.

WG1 Ballot Items To Finalize By Sep. 30

WG1 - Core

#121 The semantics of expt for zero bases has been refined

The R5RS definition of expt is:

-- procedure: expt z1 z2 Returns Z1 raised to the power Z2. For z_1 ~= 0 z_1^z_2 = e^z_2 log z_1 0^z is 1 if z = 0 and 0 otherwise.

however exponents with negative real parts are undefined. R6RS attempted to clarify this with:

0.0^z is 1.0 if z = 0.0, and 0.0 if (real-part z) is positive. For other cases in which the first argument is zero, either an error is signalled or an unspecified number is returned.

(Ignore the change in exactness, which was strictly editorial and the examples clarify that the rules ignore exactness.)

This is unique in all the reports of a result either signalling an error or returning a value. The motivation for this was because R6RS consistently removed uses of the "is an error" terminology which would more naturally fit this situation.

An alternative, r5rs-error, is to restore the "is an error" text since we are not avoiding this in R7RS:

The value of 0^z is 1 if (zero? z), 0 if (real-part z) is positive, and an error otherwise. Similarly for 0.0^z, with inexact results.

The /real variant restricts the domain for the zero base exception to the real numbers. This is because 0z is mathematically undefined for non-real z, and implementations do not agree on the result.

I agree that the R6RS rule makes no sense in an R7RS context. However, it's worth saying explicitly that the oddball zero cases are errors.
This seems consistent with #367. According to Wikipedia, for pos real b, bc = ecln(b) (the parens may be missing in the R5RS snippet?). The zero base, non-real exponent case can be defined to return nans and we should not preclude that.
R7RS isn't making the is-an-error change. I'm choosing "/real" over non-"/real" because there isn't enough agreement to support the latter.
As I understand the above text is just false: 00 and 0.00.0 are mathematicaly undefined, this is because it is not continuous there. Just take x(-1/log(x)), when x -> 0 it is equal (and therefore converges) to 1/e instead of 1 ! Provided this is changed I prefer the openness of r5rs-error. Bradley's argument convinces me to retain 00=1 (i.e. only if we have an exact 0 as exponent) as a practical convention.
The entire rationale for R6RS not using this option doesn't apply to R7RS.
The R6RS approach isn't applicable, and I prefer explicit errors.

#472 clarify semantics of non-library library declarations

In items #91, #148 and #150 we voted to allow the use of include, include-ci and cond-expand at the "top-level" respectively, but there remains some confusion as to their semantics.

Here "top-level" refers to repl and program body top-levels, but not library bodies.

One interpretation is that these behave like library declarations, and can expand into import forms. In this case, for a purely static implementation of R7RS libraries, they must first be statically scanned from all top-level forms. They cannot be used outside the top-level, and are not even available as bindings otherwise. This is the declaration proposal.

Another interpretation is that they are just normal macros with the obvious definitions (cond-expand in terms of the output of the features macro), are available in the (scheme base) library, and consequently can't be used to expand into import since imports have already been resolved. This is the syntax proposal.

Alternately, we could provide both. If you think this is all too confusing you could also vote remove, to drop these extensions.

Declaration is the option that makes sense to me, without however permitting declarations in included files (they are currently forbidden). I see no reason in these cases to make a distinction between library bodies on the one hand and programs and REPLs on the other. The syntax option allows them to be used in random nested places, which I consider to be unnecessary.
I don't like the idea of forms being "inherently" top-level only.
There's just too much confusion in this area.
These are common and useful forms, but having them as a separate declaration form, especially for include and the like, is very confusing IMO, especially for implementations that will choose to provide a syntactic include nonetheless.
With the confusion I'd just as soon remove these. If we're going to have it, it's more useful as syntax (as the original commenter wanted), and it encourages better encapsulation to force declarations into libraries.
"declaration" seems the simplest. "both" seems the most complex. "remove" seems to be throwing the baby out with the bathwater.

#473 library declaration locations in top-level

R6RS allows only a single library declaration, import, at the beginning of a program body, and this must contain all imported libraries.

Pending the result of ticket #472 we may also allow include(-ci) and cond-expand to expand into imports, and so the single form restriction would not make sense. However, it would be reasonable to restrict all library declarations to the beginning of a program - the first non-declaration would begin the real body. This is the beginning-only option.

The advantage of the r6rs proposal is that it would not require any changes in existing R6RS program loading implementations to support. If the result of ticket #472 indicates multiple declaration types are available this option would automatically become invalid, so you don't need to vote against it on those grounds.

The advantage of the beginning-only option is that it becomes possible to statically determine all program imports without expansion, which was the primary motivation of a static library system.

The final alternative is any-top-level, which allows these forms anywhere at the top-level, possibly interspersed with definitions. The advantage of this is that you can cut&paste repl sessions (for which interspersed imports are always allowed) as a program. The disadvantage is that programs can no longer be resolved separately from expansion.

Note that this is about programs only, not REPLs or library bodies. I really, really dislike both any-top-level and beginning-only. The first is too flexible, the second, not flexible enough. Very reluctantly I choose beginning-only because it preserves static analysis. I see no benefit to the r6rs option at all, given that R6RS systems will have to provide additional support for R7RS library syntax anyway.
I think that import should generally act like a multi-define, and so should be usable like a top-level define. The question of redefining import is a separate one, and should be discussed separately.
As long as we're only restricting what the standard supports but are not restricting how implementations may extend their own implementations, I'm fine with this. In that case, preserving R6RS compatibility is a good idea.
If applicable we should strive for at least this much compatibility with R6RS. Otherwise, we definitely should not allow any-top-level which defeats the purpose of having a static library system.
beginning-only is a conservative minimum to require; implementations might choose to be more flexible without becoming incompatible.

#405 Retract language requiring force to accept non-promises

#405 lumped together several issues, one of which was a requirement (as opposed to an option) to make force applied to a non-promise return its argument, as opposed to it being an error. Thus, it would require (force 2) => 2. However, R6RS requires (force 2) to signal an error, and many non-R6RS Schemes also signal an error (see ForceNonPromise for details). These facts were not considered at the time.

Vote retain to retain this requirement, or retract to retract it and leave the result of (force 2) implementation-dependent.

I can't see forcing all R6RS systems into non-compliance over this small point.
If a programmer needs to know what is and is not a suspension before forcing it, suspensions are not that different from thunks (so why bother). It should be possible for a portable program to be lazy (sorry) and not have to worry about whether something is a suspension or not. This requirement does not break any programs, and there is no other reasonable value to return. Also, extending forcing in this way seems consistent with the implicit forcing that occurs on primitive application.
There isn't enough agreement among implementations to impose the new requirement.
This was just an oversight when the item was originally proposed - there's no grounds to require this.