P1370 Generic numerical algorithm development with(out) numeric_limits

[jensmaurer]

P1370R0 Generic numerical algorithm development with(out) numeric_limits (Mark Hoemmen, Damien Lebrun-Grandie)

[Lawrence-Crowl]

SG6 in Kona
Revise with additional 'safe_min' value, name subject to future bikeshedding.
SF 5 F 0 N 2 A 0 SA 0

Revise with explanation of the distinction between the various constants.
SF 5 F 2 N 0 A 0 SA 0

Merge into a revision of P0437R2.
SF 5 F 2 N 0 A 0 SA 0

[mhoemmen]

Would y'all happen to have the LEWG-I minutes? I didn't see them posted on the wiki. I'd like to get the polls right for the feedback section in the post-meeting mailing update.

[wg21bot]

P1370R1 Generic numerical algorithm development with(out) numeric_limits (Mark Hoemmen, Damien Lebrun-Grandie)

[brycelelbach]

Kona 2019-02 LEWGI Minutes

P1370R0 Granular std::numeric_limits<T>::min

Champion: Mark Hoemenn

Minute Taker: Steve Downey

Start Overview: 15:45

Start Polling: 16:05

POLL: We want a boolean type trait has_tiny_v<T> and tiny_v<T> is not meaningful (but not SFINAEd away) if has_tiny_v<T> is false; likewise for safe_minimum_v<T>.

Strongly For	Weakly For	Neutral	Weakly Against	Strongly Against
4	2	1	1	0

Attendance: 11

That has consensus.

POLL: We should call tiny_v<T> (smallest normalized positive floating point value). Vote for as many as you like:

Options	Votes
`tiny_v (term of art in Fortran)	3
minimum_positive_v	0
min_pos_v	0
minimum_normalized_v	0
min_normalized_v	3
min_normal_v (term of art in IEEE 754)	7
underflow_threshold_v	1

Attendance: 12

min_normal_v has consensus.

POLL: We should call safe_minimum_v<T> (smallest number such that 1 divided by this number does not overflow). Vote for as many as you like:

Options	Votes
sfmin_v (term of art in LAPACK)	2
safe_min_v	0
safe_minimum_v	0
reciprocal_overflow_threshold_v	6
rot_v	0
min_invertible_v	9
min_divisor_v	0

Attendance: 12

min_invertible_v has consensus.

End: 16:10

CONSENSUS: Bring a revision of P1370R0, with the guidance below, to SG6/LEWGI for further design review.

• Rename tiny_v to min_normal_v.
• Rename safe_minimum_v to min_invertible_v.
• Change min_normal_v and min_invertible_v to always be defined not SFINAE'd away, add boolean has_min_normal_v and has_min_invertible_v traits, and make the value of min_normal_v and min_invertible_v only be meaningful when the corresponding has_* trait is true.

[mhoemmen]

@brycelelbach I may not have had the right poll results at submission time. I can't make it to Cologne but Damien or one of the Sandians should be able to attend any SG6 or LEWGI review.

[brycelelbach]

Cologne 2019-07 LEWGI Minutes

P1370R1 Granular std::numeric_limits: Design Review

Champion: Damien Lebrun-Grandie

Minute Taker: Tobias Lowe

Start Overview: 07-16 8:44

Start Review: 8:52

CONSENSUS: Feedback from last LEWGI review was not entirely addressed because it needs to be applied to another paper, not this one. LEWGI may need to discuss the related paper, P0437.

[mhoemmen]

I look forward to seeing the minutes. Note the following part of R1:

The LEWG-I minutes were not posted, but the authors remember the following discussion points.

Did the paper totally get this wrong?