lewg-minutes.txt

<2019-02-21 Thu 13:05>
Intro to Net::TS

ChrisK: presents

basics

experimental/executors for async IO

experimental/io_context

experimental/timer

experimental/buffer

Bryce: thinking about buffers : untyped data(?)

experimental/socket
stream, datagram, ....
socket options
iostreams

experimental/internet
IP addresses & network
endpoints
name res
TCP, UDP
inet socket options

async:
io_context is like select/poll/completion port

get resultt via std::future

use fibers



<2019-02-21 Thu 13:22>


p1100r0

Chris: attempts to solve problem with composing dynmaic buffers

Lewis: This is a concept?

Chris: Yes
example wrapping a std::vector reading 10 chars and data grows
read_until, buffer filled by OS, return size is how much to use. data.erase(0, n) // loop on read_util

dynamic_buffer, buffer wrap types for different uses

slide with interface
Bryce:: what does ...
CK : next slide

readable vs writable bytes. There's a split like iostream
Bryce: Why does iostream do that

Jonathan Wakely: String stream , two high water marks, bytes read and written

Problem: Moved from state for DynamicBuffer as a wrapper

B: Why is that a problem
Z: When you want to control the state of what you read inherent to poblem

JW: library layering means abstracting the buffer, but the vector is concrete

expose abstartion like http protocol, not using vector


CK: Proposed: dynamc_string_buffer with a release() method


Counter Prop:DB has two resps. 1) grow shrink 2) track read and write
resp moves from DB to alg using it.
Copy rath than Move
just bytes, no read or writable
adds resize
B: why is this CopyConstructible
Z: cheap to copy, and

Kirk: More complicated add DB(DB&), users must track bytes, harder to use
M:

Kirk: add more interface to slice data

G When to use commit a problem because tracking state

CK: commit because the read and write get mixed

Kirk: pop_front_in

CK: consitered

LB:

JW: DynB gives you mutable buffer

CK: exploer names

Br: shrink vs resize

LB: Old interface allowd defrer shifting down. Consume may move in bytes by shifting pointers


CK: circular buffers better

Brian: trade off, safety and facility

Bryce: all interfaces haver some hazzard

Z : used by library
CK: not used by users, yes
Z: library authers have to deal
Kirk: this looks like a container

Br: string_view of span

Z: alternate is a tag type, which is terrible

CK: examples Almost no change for user code

Br: nice cheap to copy handle vs


Kirk: DBRef type is just moveable.

SD: Compatiblity concerns, migration and compat with asio and beast

<2019-02-21 Thu 14:06>
Buffer Sequence
ck: multiple heterogenous buffers into or from buffers

SD: Ranges are not heterog
Z: how does this work
Does Range::concat do this?

If concat does this do we want that?

CK: Convertiblity and const-ness
Ryan: Scatter API.
Fund usecase
G library headers and user headers all get emitted, lifetime problems keeping alive until sent on network.


Discussing if we've misread level: result, yes

buffer_cat creates a sequence of buffers from a heterog collection of buffers


Bryce: what parts are network and generic , look at ranges

JW: We will look at ranges but problems with void*



<2019-02-21 Thu 14:32>


Associations for Call Wrappers

Type adapters for std::bind and std::ref and LEWG doesn't want to touch bind.


<2019-02-21 Thu 14:42>
Enahancements to enable customs iO executors

Motivation: MS would like executor to run on system thread pool rather than donated threads

easier for users to specify executors

Far more than just a thread pool

Br: This is not Executors, just net executors
CK: Yes
Br: make that clear

CK: In Boost master

Z: does this affect coupling with Executor TS
CK: No. Interaction is largely the same

L: Can users provide your own implememtation
CK: Not readily
B: With Executors could they
CK: That's a large piece of work,
L: how does the gneric work with inheritance
CK: a bag of types indexed by kind of service, can then be created by cliets in ther context

Should have been done earlier


<2019-02-21 Thu 15:23>
Hand off scribe



<2019-02-21 Thu 15:44>
	Granular std::numeric_limits

Please ask Lawrence Crowl that third poll does not mean what it says

Need names for TINY and SFMIN, TINY from fortran and safe min from lapack

smallest normalized floating point number
is optional or required ? sfinae out
also apply to Walters' paper to break out numeric traits

JF: no sfnae traits  has_tiny and a value is always for tiny_v

Mark: are you saying my people can't write C++

JF: No one should have to write detection algorithms, and yes.
And the call what they write 'codes'


Mark: sg6 asked for bikeshed

B: SAFE is a bad bad name and LEWG would kick it back with prejudice
<2019-02-21 Thu 16:12>

Hand back to Mark