[cfp-interest 4011] JTC1/SC22/WG14/CFP 2026/09/30
Jerome Coonen
jcoonen at gmail.com
Wed Sep 30 12:52:16 PDT 2026
CFP Meeting
2026/09/30: 8:00 AM PDT/3:00 PM UTC
[Please submit proposed changes to these minutes to Jerome or to the group.
Revision changes appear at the bottom.]
Attendees and introductions
Rajan Bhakta, Jerome Coonen, Damian McGuckin, Ariel Burton,
Joshua Cranmer, David Hough, Tue Ly
Agenda
https://cfp-wiki.esi.com.au/pub/CFP/WebHome/n3940.pdf
Previous meeting notes
https://cfp-wiki.esi.com.au/pub/CFP/WebHome/260826-Notes.pdf - 2026/08/26
Minutes
Study group logistics
Next meeting: 28 October 2026, 8:00 AM PDT/3:00 PM UTC
Note: Zoom link may change for this meeting or in the future. Email
update will be provided if it is for this meeting.
C documents
The latest C2Y draft is N3886 2026/05/24:
https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3886.pdf
C23 has been published ISO/IEC 9899, available for purchase.
https://www.iso.org/standard/82075.html
IEEE 754 update
Damian: Discussion of the refining language of NaNs, divide by zero, and
other extraordinary issues. There is also more work around correct rounding.
Jerome & group: We have looked at the objectionable use of “pole” to
describe all instances of division by zero. We hope to improve our wording
based on 60559.
WG14 update
Rajan: Finished Aug meeting. Next meeting 2026/11/16-21 in Brazil, deadline
a month before.
C++ liaison
Joshua: Next meeting coincident with C meeting in Brazil. Early work on an
“Annex F” for C++. SP22, compatibility group, will also meet. Ordinarily,
these meetings are offset by a week,permitting attendance at both, but in
this case the site is available for just one week.
TS-4 and TS-5 revisions
None
News
Jerome: Travel necessitates quick turn-around of these minutes. Comments to
Jerome or CFP will be addressed after 7 Oct.
Carryover action items from last meeting
Fred: Add to open items list the issue in 3928 from Trevor regarding the
awkward language in H.4.1 regarding whether or not functions need to be
supported, based on what types are supported.
Carry over
Group: Look at the notion of “range of representable values”. Can we
tighten up this usage?
Done
Action items from last meeting
Jerome & Rajan: Fix and submit proposal for improved wording in H.11.4.1.
Done n3962
Fred: Submit the nextafter/nexttoward proposal in email 3963.
Carry over
Ariel: Investigate the sign of fromfp(-0.0) across IBM compilers. Group
invited to check other systems.
Ariel: Looked at nextafter(). Behavior nextafter(-0, +0).
Joshua: glib and llvm libc have fromfp, most other libraries don’t. Not
Intel nor others.
Jerome: Gives us freedom to do right thing. Jim Fred
carry over (enhanced by spontaneous discussion).
Fred & group: With input from the group, expand rationale and discussion of
nextafter and zero cases
Carry over
Group: Interested members submit input on Fred’s email 3964 re. nextafter()
behavior around zero.
Carry over
Group: Submit thoughts about how SNaNs ought to behave in the printf/scanf
functions.
Carry over, after some inevitable review of the issue and its ties to
function calling conventions.
Jerome & Rajan: Send email to Paul (who is probably reading these minutes
8^) ) re. signgam as a POSIX issue.
Done cfp3998
Discussion of issues (as of 2026/09/28)
-
[cfp-interest 3962] Quantum of fromfp - Fred Tydeman - For reference
(completed item)
Carry over
-
[cfp-interest 3966] SNaN not raise invalid - Fred Tydeman
Carry over
-
[cfp-interest 3967] Annex F - Introduction + SpecialCases - Damian
McGuckin
Damian: Not planning changes other tahn F.2.2 on “representable values”.
Rajan: There’s no need to delay on “range of representable values”
issue, which will take time to resolve.
Action
-
[cfp-interest 3969] Re: binary vs decimal sNaN initialization - Rajan
Bhakta
Rajan: Fred has completed and I’ve reviewed but we’ll carry over.
-
[cfp-interest 3965] Model numbers and double-double - Fred Tydeman
Carry over
Ariel & group: What are the model numbers? See 5.3.5.3.3 Characteristics
of floating-point types <float.h> for the concepts behind “model
floating-point numbers”.
Ly: We are looking more at doubledouble and Nvidia’s floatfloat. The
context is llvm libc. For performance, we prefer doubledouble to float128
for high-precision needs.
-
[cfp-interest 4006] Action: range of representble values, revisited
<https://mailman.oakapple.net/pipermail/cfp-interest/2026-September/003988.html>
Jerome Coonen
Rajan & Jerome: Table this proposal and focus on the discussion in the
next item.
Closed
-
[cfp-interest 4008] Re: Action: range of representable numbers
<https://mailman.oakapple.net/pipermail/cfp-interest/2026-September/003990.html>
Jerome Coonen
This discussion dominated the technical conversation. Given the
note-takers active role in it, this summary is a best effort to capture key
observations. Most important is the new action item below.
Jerome: Reviews the several uses of range of representable values
(RORV) that pertain to floating point. Reiterates the interpretation that
RORV refers to just the values of computer arithmetic
Joshua: Another interpretation, and the one taken by C++, says that
this suggests that the value 0.1 is not a valid constant because it is not
exactly representable. One section just cited is a constraint that requires
that a compiler error be issued if the constant is no in the RORV.
Jerome: It’s more subtle to talk (very precisely) about just which
values may “project” onto a floating point number system. In my view, there
is no need to do so. The point is that there is a mechanism to convert the
text expressions of numeric values into computer numbers.
Joshua: Here's the C++ wording for constexpr evaluation:
https://eel.is/c++draft/expr.const.core. C++ is based on C, and it may
have been unclear what was meant by RORV when the C++ concepts were
elaborated. Now, infinities and NaNs are not considered part of the RORV.
Jerome: The whole point of 754 forty-five years ago was to avoid just
such “errors” by providing infinity and NaN symbols to ensure that there
would always be a representable value at the end of a computation.
Joshua & group: C and C++ have different rules, so we need to sort that
out. The biggest computational issue may be whether extraordinary
expressions like, “1.0/0.0” or “0.0/0.0” or “3.0e5000000000” or even
“1.0e500 (overflows in double)” can be permitted in constants. In a 754
environment, there is a possible representable value. Possible questions
are what is intended for the compiler and what is the best language to
realize that?
Other issues
-
https://cfp-wiki.esi.com.au/pub/CFP/WebHome/c26h.pdf
Rajan: As a note to readers, these items are referenced in discussion by
their WG14 doc numbers or CFP email numbers, not by their index numbers in
this list.
Adjournment
9:40 AM PDT
-----------------------------------------------
Action items to be carried over
Fred: Add to open items list the issue in 3928 from Trevor regarding the
awkward language in H.4.1 regarding whether or not functions need to be
supported, based on what types are supported.
Fred: Submit the nextafter/nexttoward proposal in email 3963.
Ariel & group: Investigate the sign of fromfp(-0.0) across IBM compilers
and other systems.
Fred & group: With input from the group, expand rationale and discussion of
nextafter and zero cases
Group: Interested members submit input on Fred’s email 3964 re. nextafter()
behavior around zero.
Group: Submit thoughts about how SNaNs ought to behave in the printf/scanf
functions.
New action items
Damian & Rajan: Damian to send clean copies of Annex F Intro & Special
Cases to Rajan, who will submit to WG14.
Jerome & Joshua & group: Research the use of “range of representable
values”, what its definition might be, how the term itself might be
rephrased, and its ties to C++ and to related C concepts like model numbers.
Carryover discussion items
-
[cfp-interest 3962] Quantum of fromfp - Fred Tydeman - For reference
(completed item)
-
[cfp-interest 3966] SNaN not raise invalid - Fred Tydeman
-
[cfp-interest 3969] Re: binary vs decimal sNaN initialization - Rajan
Bhakta
-
[cfp-interest 3965] Model numbers and double-double - Fred Tydeman
Carry over
Signoff
Respectfully submitted.
-Jerome Coonen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20260930/c871f91d/attachment-0001.htm>
More information about the cfp-interest
mailing list