1 2 3## Agenda for the March 26 video call of WASI Subgroup 4 5- **Where**: zoom.us 6- **When**: March 26, 16:00-17:00 UTC 7- **Location**: *link on calendar invite* 8- **Contact**: 9 - Name: Dan Gohman 10 - Email: sunfish@mozilla.com 11 12### Registration 13 14None required if you've attended before. Email Dan Gohman to sign up if it's 15your first time. The meeting is open to CG members only. 16 17## Logistics 18 19The meeting will be on a zoom.us video conference. 20Installation is required, see the calendar invite. 21 22## Agenda items 23 241. Opening, welcome and roll call 25 1. Please help add your name to the meeting notes. 26 1. Please help take notes. 27 1. Thanks! 281. Proposals and discussions 29 1. WASI currently uses a POSIX-like `errno` enum, but that seems 30 not well suited for less POSIX-like APIs: 31 https://github.com/WebAssembly/WASI-crypto/pull/13#issuecomment-599324384 32 Many systems have evolved to having a single error enum shared 33 across the whole system. Is it ok if we don't aim for that? 34 1. POSIX tries hard not to let syscalls cause segfaults, returning 35 things like `EFAULT` instead. There may be some value in this for 36 POSIX compatibility, but how valuable is this otherwise? Should 37 other WASI APIs be allowed to trap? 38 1. Evolving idea: wasi-sdk API for extra emulation: 39 1. `-D_WASILIBC_EMULATED_MMAN` 40 1. `-D_WASILIBC_EMULATED_PAUSE` 41 1. `-D_WASILIBC_EMULATED_RAISE` 42 1. ... 43 1. `-D_WASILIBC_EMULATED_ALL` ? 44 and 45 1. `-lwasi-emulated-mman` 46 1. `-lwasi-emulated-pause` 47 1. `-lwasi-emulated-raise` 48 1. ... 49 1. `-lwasi-emulated-all` ? 50 1. [Proxy-Wasm](https://github.com/proxy-wasm) update. 51 52## Meeting Notes 53 54Attendees: 55 56Dan Gohman 57Jan Falkin 58Lee Campbell 59Till Schneidereit 60Alex Crichton 61Andrew Brown 62Mingqiu Sun 63Olaf Tomalka 64Piotr Sikora 65Pat Hickey 66Peter Huene 67Sam Clegg 68Johnnie Birch 69Wouter van Oortmerssen 70Luke Wagner 71Aaron Turner 72Benjamin Brittain 73Taylor Thomas 74John Plevyak 75Ralph Squillace 76Brian Hardock 77 78Meeting notes: 79 80WASI currently uses a POSIX-like errno enum, but that seems not well suited for less POSIX-like APIs: https://github.com/WebAssembly/WASI-crypto/pull/13#issuecomment-599324384 Many systems have evolved to having a single error enum shared across the whole system. Is it ok if we don't aim for that? 81Lee: we need standardized ways to propagate errors throughout the system which is generic over all kinds of APIs 82PH: Design a way to structure and compose errors. We should do this with types, rather than integer ranges. 83SBC: Capability system is one of the unifying parts of WASI, there will be errors common to all capability operations. 84PCH: Compositional errors might be something like a sum type of different errors. 85Bwb: Strings as errors have lots of awkward problems, including internationalization. 86PCH: Being able to use any type gives us the power to describe errors. 87Jplev: Have a generic error codes, with ways to map to them. 88Sbc: Don’t redeclare the same errors as POSIX 89Pat: APIs define the things they need now, and we’ll work on factoring things out and composition as we go forward. 90Lee: POSIX is a C API and we can map WASI errors to POSIX errno as needed. 91 92 93POSIX tries hard not to let syscalls cause segfaults, returning things like EFAULT instead. There may be some value in this for POSIX compatibility, but how valuable is this otherwise? Should other WASI APIs be allowed to trap? 94Sbc: one of our objectives is to be able to implement WASI as wasm modules. 95Lee, sbc: We seem to need core language support for catching traps to really figure this out. 96Evolving idea: wasi-sdk API for extra emulation: 97``` 98-D_WASILIBC_EMULATED_MMAN 99-D_WASILIBC_EMULATED_PAUSE 100-D_WASILIBC_EMULATED_RAISE 101... 102-D_WASILIBC_EMULATED_ALL ? and 103-lwasi-emulated-mman 104-lwasi-emulated-pause 105-lwasi-emulated-raise 106... 107-lwasi-emulated-all ? 108``` 109Lee: Do we need the `-l`s? We can just put everythgin in libc.a and it’ll only get linked in 110Proxy-Wasm update. 111This is an update from the previous presentation in October. 112Consolidated many callbacks from the previous presentationtation. 113 114Sbc: Have you looked at creating a witx description? 115Piotr: Not yet; we’re still iterating on the API. 116Sbc: Do proxy implementations need to implement all callbacks, or are some of the m optional? 117Piotr: Some are optional. If you’re only implementing HTTP, you only need to impelment HTTP callbacks. 118PH: This proposal is great, and also very big, can we factor out pieces that can be used in other contexts? 119 120Adapting the core CG Phases for WASI. 121https://github.com/WebAssembly/proposals/ 122“Two or more Web VMs” 123 124Reference interpreter, formal semantics 125 126Fitting the existing phases into the process 127
