The README states that V was to be released on 2019-06-22T18:00Z but it is 2019-06-22T18:06Z and no release exists

edit: it's 18:24Z and a release exists, ty

I downloaded the compiler source and IcarusVerilog couldn't process it but there are no other compilers available that can process .v files, please fix

Very interested in your input.

https://github.com/vlang-io/V/blob/master/examples/users.v https://github.com/vlang-io/V/blob/master/examples/hello_world_gui.v https://github.com/vlang-io/V/blob/master/examples/generic_repository.v

Transferred from a document posted here in these documents by @cristian-ilies-vasile:

V concurrency high level design

After a carefully consideration of proposed concurrency models and suggestions expressed on discord v-chat channel the high level design is based on GO language model (message passing via channels) which in turn is a variation of communicating sequential processes.

I did read papers on actor model but seems that coders do not use all the primitives provided by language and resort to threads and queues (see Why Do Scala Developers Mix the Actor Model paper).

Almost all high level requirements are taken from Proper support for distributed computing, parallelism and concurrency published on github.

Because there are many words used more or less interchangeably like coroutine, goroutine, fiber, green threads etc I will use the term green thread.

Key points

• the language will provide primitives able to deal with green threads – the green threads monitor/scheduler will be part of the language
• there will be no yield() points in code; the scheduler can suspend / resume / block / cancel any green thread in a preemtive mode
• based on memory model the green thread will be stackless or stackful
• one green thread could spawn n others green threads
• a green thread must be reentrant
• the scheduler could start a new instance of a green thread in order to accommodate load bursts (elastic computing)
• Green threads communicate using message passing (channels) instead of shared variables

Open points

• mux/demux green threads on real threads how to?
• structured concurrency
  • Martin Sústrik - http://250bpm.com/blog:71
  • Roman Elizarov - https://medium.com/@elizarov/structured-concurrency-722d765aa952
  • Alan Bateman - https://wiki.openjdk.java.net/display/loom/Structured+Concurrency
• depending on application scope could have cpu bound or io bound or both type of green threads; do we need 2 different approaches?
• Debugging
• how to catch programs bugs (Heisenbugs) before the code is shipped to prod systems?
• Formal methods / TLA+

There is one thing in V I do care a lot about - parallelism & concurrency for seamless local and distributed programming.

Note, I didn't want to write this to https://github.com/vlang/v/issues/561 as the scope I'm describing is broader.

I'm confident every new language must provide first-class support for both parallelism and concurrency. Concurrency to allow one physical CPU thread do cooperative scheduling and parallelism to allow spawning the same go/v routine on several physical threads. This is actually what go lang does (when GOMAXPROCS is set to 2 or more and the underlying operating system supports threads) except for always spawning just one go routine disregarding the number of physical CPU cores (which is where V should improve on - see below).

Why both concurrency and parallelism and not either of them?

In the world of big.LITTLE, GPU/FPGA offloading, CPUs with tens/hundreds of physical cores (and hyperthreading on top), mobile devices trying to save as much energy as possible, NUMA supercomputers, mobile & close vicinity networks versus optical fiber networks etc. it's inevitable to dynamically schedule computation directly in runtime of each application (not just operating system wide which is too coarse and thus inefficient and not just cooperatively which would use just one physical CPU core per application which makes sense only for power saving scenarios, but nowhere else).

We have instruction-level paralellism covered (it's not yet present in V - see e.g. my rant about restrict - but it's a "solved problem" in todays compilers). The rest of the parallelism is just about "how close can we get to pure dataflow programming". Which appears to be kind of difficult.

Because instruction-level paralelism is solved, the best approach nowadays seems to be to:

1. write as much serial code which allows being highly optimized by instruction-level parallelism at once in one go/v routine (typically a tight loop or one tree of nested tight loops or a slow I/O or offloading to GPU/FPGA or similar) - the outermost construct of this go routine will be always an endless loop yielding (calling the cooperative/parallel scheduler) every N iterations

2. then schedule these go/v routines in cooperative manner on one physical CPU core with fixed-size queue (single-producer-single-consumer "SPSC" - for performant implementations see discussion below) in between these go/v routines (this queue is called a channel in go lang); the queue size might be between the size of L1 and 2*L2 cache as suggested in paragraph 3.3 in Analyzing Efficient Stream Processing on Modern Hardware.

3. and if any go/v routine shall become a bottleneck during computation (i.e. consumer is slower than producer for some time - this can be easily monitored by the cooperative/parallel scheduler as metric "how full are the queues"), a new instance of the slow consumer go/v routine shall be spawned (with respect to reentrancy) in a different operating system thread (i.e. on a different physical CPU core) including all the plumbing work (e.g. spawning an additional go routine acting as multiplexer getting the producers outgoing data and multiplexing it to the original as well as the new instance of the consumer; and spawning a demultiplexer go routine to join the outgoing data from both consumers) - this everything would the cooperative/parallel scheduler do.

4. and if the queues shall become almost empty for some time, remove the multiplexers and demultiplexers.

The cooperative/parallel scheduler as referenced above shall be a user-definable routine (if not present, a default built-in scheduler working similarly like described above will be used). This scheduler shall run preemptively (in its own thread?) and get as input arguments at least the following: thread handles, pointers to all go/v routines and corresponding queues between them, pointer to the built-in scheduler (in case the user just wanted to wrap it), and pointer to user-defined metadata (e.g. statistics allowing better scheduling judgement). This would allow for really complex scheduling on embedded systems as well as NUMA systems or any other highly dynamic systems (e.g. a smartphone could leverage being shortly connected to a cluster of other computers or smartphones and offload some slow algorithm there etc.). See e.g. MultiQueues.

This way one can write application once and deploy it to your grandmas wrist watches as well as to a supercomputer (assuming the user-level scheduler is a "bit more" clever than outlined above - imagine complexity similar to an SQL query optimizer for a distributed DB). It's also a highly failure-tolerant system (imagine Erlang which supports even an online update of the whole application in memory while serving all clients under full load without interruption!). Also as you may have noticed, go lang does not scale first because there is no spawning of redundant go routines (those which will be bottle necks) to other physical CPU cores and second because go doesn't have a dynamic user-influencable scheduler (taking into account advanced statistics, power consumption, etc.).

There is one catch though. If implemented naively, then such elasticity doesn't guarantee global ordering among channels/queues (ordering works only inside of one channel) and real life scenarios in IT are unfortunately more often than not relying on ordering. This has to be accounted for and will require user intervention (i.e. be syntactically explicit). Either in the form of knowledge where to (a) insert "tagging" of items before they'll be spilled among different channels and (b) where the "tagging" will be removed and used to assemble the original ordering. Or in the form of assigning a strict priority to each channel. Or using other scheme.

See also lock-free and wait-free datastructures (state of the art as of now) which might be handy for an efficient implementation of parallelism.

IMHO we could actually postpone the implementation of concurrency and stick with just parallelism (as it's easier to implement than the interleaving between concurrency and parallelism as outlined above), implement the queue and a basic scheduler and first then (maybe even after V 1.0) get back to concurrency. As of now we have parallelism without queues and without the scheduler, so we actually already have a good starting point.

Table of parallelism possibilities we have nowadays (just for reference):

parallelism kind | suited for execution duration | latency overhead | suited for how frequent execution startup | internode bandwidth quality has influence | special parallel SW architecture required | requirements for execution & deployment ---| ---| ---| ---| ---| ---| --- CPU non-JIT vectorization on a CPU core | very short to long | ~none | frequent | none (it's just 1 node) | no | binary for the particular CPU CPU JIT vectorization on a CPU core | short to long | low | moderate to frequent | none (it's just 1 node) | no | app source code + VM binary for the particular CPU CPU cores utilization (thread, process, ...) | long | low to moderate | moderate | lower to none (it's just 1 node) | yes (except: pure objects as actors or alike) | binary for the particular CPU accelerators (GPU, FPGA, ...) | long | low to moderate | moderate | lower to none (it's just 1 node) | yes (except: array/matrix/tensor/...-based languages) | binary for the particular CPU and accelerator (GPU, FPGA, ...) supercomputer with NUMA topology | long | moderate | sporadic to moderate | moderate (hypercube/... is really fast) | yes (except: pure objects as actors, etc., array/matrix/tensor/...-based languages) | binary for the particular CPU and accelerator (GPU, FPGA, ...) LAN cluster | long | moderate to high | sporadic to moderate | moderate to high (can be fast, but not always) | yes | binary for at least 2 CPUs and/or at least 2 accelerators (GPU, FPGA, ...) WAN cluster | long | high | sporadic | high (basically can't be that fast) | yes | binary for at least 2 CPUs and/or at least 2 accelerators (GPU, FPGA, ...)

(in practice these might overlap and/or be used simultaneously)

P.S. The dynamic scaling over the different kinds of parallelism as outlined in the above table is called "fine grained distributed computing". And if anything from the above proposal sounds crazy to you, then I can assure you, that the world doesn't sleep and there is at least one seamless (fully dynamic) solution offering first class fine grained distributed computing - Unison.

Other things to consider and/or track:

1. As of March 2021 the fastest publicly known (and most comprehensive & general) parallelism & concurrency backend/library is Weave
2. use PRNG (pseudo-random number generator) which copes well with dynamic threading (i.e. doesn't suffer from "same seed leads to same random numbers in all threads") - see e.g. splittable PRNGs such as JAX
3. maybe infinite loop in rand https://github.com/vlang/v/commit/a7c84834f4ba4948b8102a05b603bf8d51484760#r39632493
4. g_str_buf with parallel access:
   • https://github.com/vlang/v/commit/778a1cc34ae5674c345d73e2a507bce9f77dea1b#commitcomment-36848204
   • https://github.com/vlang/v/commit/0f92800921c706de0bd872f7eff30047a99c734d
5. FPU stuff needs special attention from the threading (work stealing) scheduler - see https://github.com/mratsim/weave/issues/163
6. under the hood, we might utilize some wait-free and lock-free algorithms (see e.g. https://github.com/pramalhe/ConcurrencyFreaks )
7. reconsider deadlock/livelock/no_thread_running/... detection - see https://github.com/vlang/v/issues/10340
8. incorporate a "sharded" RwMutex into the standard library and use it also for V's built-in "sharded" data structures (map etc.)
9. memory models of multicore architectures (x86, ARM, POWER, ...) are still sometimes not fully formally defined but SW memory models are even worse at that :open_mouth: https://research.swtch.com/mm

Current:

mut m := sync.WaitGroup{}
a := m // m is duplicated
z := someStruct{wg: m} // m is duplicated and requires &, trap for new players
someFunc(m) // &m is passed

Suggested:

mut m := sync.WaitGroup{} // m = &sync.WaitGroup
a := m // a and m share same data
z := someStruct{wg: m} // z.wg and m share same data
someFunc(m) // &m is passed

I would suggest that by making all muts act as if they were pointers and values otherwise. This would reduce cognitive load and code. The ptr "&" syntax could be eliminated (mostly). This might help towards #41.

Thoughts?

Before we had a problem that calling foo() works even it's not wanted.

fn foo(table &ast.Table) {...}

I choosed kwargs because it's short and well known from Python.

Intention

The vweb module will be refactored to have a structure suitable for parallelization. As a community it would be best to agree on the design collectively.

The intent of this issue is to settle on a design for the refactored vweb module.

Proposal

Use vweb.Context for request specific context, and use the user supplied App struct strictly for state management. Methods defined on App will be declared using shared to enable developers to use lock and rlock as necessary.

Additionally, vweb.run will take two arguments: an initialized generic type and a vweb.Config struct. This allows developers to initialize anything they may need before server start.

Example

import vweb

struct App {
mut:
	cnt int
}

fn main() {
	conf := vweb.Config{
		port: 8082
		static_directory: '.'
		static_path: '/static'
	}
	vweb.run(App{cnt: 100}, conf)
}

['/users/:user']
fn (shared app App) user_endpoint(mut c vweb.Context, user string) vweb.Result {
    rlock {
        return c.json('{"$user": $app.cnt}')
    }
}

fn (shared app App) index(mut c vweb.Context) vweb.Result {
    lock {
        app.cnt++
    }
	// These values are available in the index.html template
	show := true
	hello := 'Hello world from vweb'
	numbers := [1, 2, 3]
	c.enable_chunked_transfer(40)
	return $vweb.html()
}

Technical Details

I haven't looked too deep into it, but I think this is not easily possible. I tried and failed making a POC passing a shared struct to a generic function, or initializing a shared generic type. So here is my question: will it be possible to do these things with V or future versions of V?

This language is not as advertised

This language does not do over 90% of what is advertised. Most of what is advertised is not coherent with most of the claims you have stated. With the new release of the playground, most of the examples in the docs fail, even many of the extremely basic ones.

• You appear to do very basic tokenization/lexing which you use to naïvely convert the code into C or C++ (through testing of the playground)
• The converter program does virtually no semantic checking other than the most basic type checking
• Name lookups do not respect nested scopes within functions and caused the playground to crash
• There is poor checking for certain characters and tokens you disallow
• Record types and their methods must be declared before use, the type before the methods
• The playground was crashing frequently
• The playground is extremely insecure

You have previously claimed that the V compiler uses no AST. However, most of the features that you claim would require a form of AST to even work, including generics and interfaces. An AST is an abstract syntax tree, which means that it just stores data about the syntax that is in a tree-like format. It's pretty much impossible to not have unless you have a very basic language or doing naïve transformations into another similar language.

You claim that the language has no runtime but the following require a runtime:

• All arrays are dynamic
• println
• Threads with go
• Synchronization of threads with runtime.wait()
• Automatic memory management

Things that are currently broken:
- Automatic imports 
- Interface method calls
- Interfaces have to be declared with `type Foo interface {` 
- Foo{a,b} syntax (use Foo{a:a, b:b}) 
- Closures
- a[i].field
- Locks (lock {}). Use sync.Mutex for now. 
- Enums 

Disabled features:
- Generics 

Why was the playground lacking a lot of basic functionality? If most of these extremely basic features are broken, how could you have made any software in this language other than basic things? -- e.g. Fibonacci examples.

Software claimed to be built in V:

• Volt
• Filey
• Vid
• gitly
• Hot code reloading

Claimed C/C++ Translations into V:

• DOOM
• DOOM 3
• LevelDB
• SQLite

Claims from your "compare" page:

- No null

How would you handle pointers?! In your documentation, you demonstrate that pointers exist. This means that no null is false. Pointers also crashed the playground.

- Much stricter vfmt

And when you mean stricter, what does that mean? It was also failing quite often even on basic code in the playground.

- Cheaper interfaces without dynamic dispatch

So how do they work? Is it just static dispatch?

What you have promised does not even come close to what is delivered.

Hi,

Update

V UI has been released, V itself is used for declarations.

I'm working on declarative UI right now.

Which approach do you think is the best: Swift/Dart (using the same language for declarations) or Qt/QML (using a smaller cleaner different language)?

It's nice to have only one language, but the "VML" approach looks cleaner to me.

Also, I'd need to modify V a lot, for example to allow things like [TextBox{}, Button{}].

The other day, I was tinkering with a small prog in V which involved enabling / disabling some of OSX's LaunchAgents. For my sins, I named one of my variables launchagentenabledpath [OK. It might not be elegant, or what you'd use, but it's how I roll!].

Anyway, to my surprise V refused to compile my prog, because it didn't like the way I'd named the variable:

bad variable name `launchagentenabledpath`
looks like you have a multi-word name without separating them with `_`
for example, use `registration_date` instead of `registrationdate`

Now, I know everyone has their own way of naming variables. Ranging from the folks who like to use cryptic one-letter names like x, to the people [like me] who tend to use VariableNamesThatAreABitMoreDescriptive.

And I know that some programming languages are fussy about what kind of names can begin with uppercase and which with lowercase letters. And then there are languages which require you to distinguish between private and public vars by [for example] beginning the names with an underscore or not. And, at the other end of the spectrum, you have a language like Nim, which is very lax about variable names, being pretty much case agnostic, apart from the initial letter.

But this is the first time I've had a programming language actually refuse to compile unless I changed a variable name –not because I'd used one that was syntactically "wrong" –but just because it didn't like how I worded it.

Is it necessary to be quite so "opinionated" as that?

Interestingly enough, I ran a few long dictionary words through the compiler as variable names [Yes, I do have too much time on my hands!] and found that, rather than any kind of logic for identifying when the miscreant programmer has created a multi-word variable name, it seems the compiler just refuses to accept any variable names longer than 15 letters.

So, ironically iamamultiword [I am a multi word] is acceptable as a variable name, whilst misunderstanding, extraterrestrial and characterisation are rejected by the compiler as being unseparated-multi-word names.

In my case, this was a minor irritation –I want to name my variables how I want [within reason]. But I could see it having a bit more of an impact for people programming in languages like German which tend to be a bit more er... 'longwinded' when it comes to naming conventions.

So, just out of curiosity, why such a relatively short limit on variable names? Did one of the dev team have a traumatic childhood incident involving one of those windowDidNotFinishLoading type names beloved of other platforms and vow never to go there again?

@medvednikov I started implementing https://github.com/vlang/v/issues/1603 by separating the "match statement" from "switch statement" in parser.v (also to fix https://github.com/vlang/v/issues/1602). I encountered the problem of parsing:

enum Color{
    red, green
}

match 1 {
    .red => println(1)
    .green => println(2)
    else => println(0)
}

We need to specify boundaries of each branch either by "new line" or by any other means. For switch V can just look on the next token to see if it's a "case" and stop statements. Your made a hack where V checks if the next token is "=>" to stop parsing statements. Also there is an option to implement error recovery to remove my limitations. I'm ready with "match statement" and "match expression". Here what I've got.

match 1 {
    else => {println(0)}
}

This code is unwrapped to

println(0)

Also you can omit those braces if you are using only one statement there.

match 1 {
    else => println(0)
}

Braces allow next case to use enums.

match 1 {
    .red => {println(1)}
    .green => println(2) // here they could be omitted
    else => println(0)
}

a := match 1 {
    else => 0
}

unwraps to

a := 0

match 1 {
    1 => println(1)
    3 => {
        println(3)
        println(4)
    }
    else => println(0)
}

This is just like switch works, but you have to use braces for multiple statements for now.

a := match 1 {
    1 => 2
    3 => {
        3
    }
    else => 5
}

match_expr requires else for now. match_expr requires the right side of the arrow to be a value. If we allow users to use statements there, then we would have to handle all cases which C doesn't support (i.e. defining variable there). This is the only difference between match_st and match_expr.

Describe the bug

The documentation contains a typo.

Expected Behavior

The correct form is:

can be written to

Current Behavior

The documentation says 2 times:

can be wrote to

Reproduction Steps

Open writer.v.

Possible Solution

No response

Additional Information/Context

No response

V version

current

Environment details (OS name and version, etc.)

any

Describe the bug

module main

import os

fn main() { config_dir := $if windows { os.home_dir() } $else { os.config_dir() or { os.home_dir() } } println(config_dir) }

Expected Behavior

Compilation ok (feature supported) or make compiler fail gracefully AND add documentation

Current Behavior

================== /tmp/v_1000/bug.403690130920050968.tmp.c:17243: error: '{' expected (got "_t2") ...

(Use v -cg to print the entire error message)

builder error:

C error. This should never happen.

Reproduction Steps

v bug.v

Possible Solution

No response

Additional Information/Context

No response

V version

V 0.3.2 a6bf20f.5daf39b

Environment details (OS name and version, etc.)

Linux Mint 20.3 64 bit

Describe the bug

Compilation failure when combining datatypes.Stack and a sumtype generic value.

Expected Behavior

I expected the sample code to compile and function correctly.

Current Behavior

/home/admin/v/vlib/datatypes/stack.v:25:20: error: cannot append `T` to `[]T`
   23 | // push adds an element to the top of the stack
   24 | pub fn (mut stack Stack[T]) push(item T) {
   25 |     stack.elements << item
      |                       ~~~~
   26 | }
   27 |
Exited with error status 1

Reproduction Steps

Smallest reproducible example:

import datatypes

type Content = string | bool

struct DataStack {
mut:
	data datatypes.Stack[Content]    
}

fn (mut dstack DataStack) push(content Content) {
    dstack.data.push(content)
}

mut dstack := DataStack{}
dstack.push('hello')

println(dstack)

Possible Solution

No response

Additional Information/Context

No response

V version

V 0.3.2 51bb630

Environment details (OS name and version, etc.)

vmodules: C:\Users\Subhomoy Haldar\.vmodules
vroot: C:\Users\Subhomoy Haldar\projects\vlang\v-main
vexe: C:\Users\Subhomoy Haldar\projects\vlang\v-main\v.exe
vexe mtime: 2023-01-01 15:13:38
is vroot writable: true
is vmodules writable: true
V full version: V 0.3.2 5acd855.51bb630

Git version: git version 2.39.0.windows.2
Git vroot status: weekly.2022.50-113-g51bb630e (1 commit(s) behind V master)
.git/config present: true
thirdparty/tcc status: thirdparty-windows-amd64 cb89a4fe

Describe the bug

I received the following error when trying to compile

==================

/tmp/v_0/main.3885071449460690037.tmp.c:7387:33: error: parameter 1 ('it') has incomplete type

 7387 | static string PGconn_str(PGconn it) { return indent_PGconn_str(it, 0);}

      |                          ~~~~~~~^~

/tmp/v_0/main.3885071449460690037.tmp.c: In function 'PGconn_str':

/tmp/v_0/main.3885071449460690037.tmp.c:7387:64: error: type of formal parameter 1 is incomplete

 7387 | static string PGconn_str(PGconn it) { return indent_PGconn_str(it, 0);}

      |                                                                ^~

/tmp/v_0/main.3885071449460690037.tmp.c: In function 'indent_pg__DB_str':

/tmp/v_0/main.3885071449460690037.tmp.c:7408:118: error: dereferencing pointer to incomplete type 'PGconn' {aka 'struct pg_conn'}

 7408 |  string _t14 = isnil(it.conn) ? _SLIT("nil") : (indent_count > 25)? _SLIT("<probably circular>") : indent_PGconn_str(*it.conn, indent_count + 1);

      |                                                                                                                      ^~~~~~~~

/tmp/v_0/main.3885071449460690037.tmp.c:7408:118: error: type of formal parameter 1 is incomplete

...

==================

(Use `v -cg` to print the entire error message)


builder error: 

==================

C error. This should never happen.


This is a compiler bug, please report it using `v bug file.v`.


https://github.com/vlang/v/issues/new/choose


You can also use #help on Discord: https://discord.gg/vlang

Reproduction Steps

vlang_simple_api.zip

I am running the code in the zip file in a docker container unzip and run the following. You can see the error in the api container

make up

Possible Solution

No response

Additional Information/Context

No response

V version

V 0.3.2 c5c7b3a.51bb630

Environment details (OS name and version, etc.)

OS: macos, macOS, 12.5.1, 21G83 Processor: 4 cpus, 64bit, little endian, Intel(R) Core(TM) i7-6567U CPU @ 3.30GHz CC version: Apple clang version 14.0.0 (clang-1400.0.29.202)

Describe the bug

examples/http_server.v does not listen on any ports.

Expected Behavior

Running telnet localhost 8080 should result in a successful connection.

Current Behavior

$ telnet localhost 8080
Trying 127.0.0.1...
telnet: connect to address 127.0.0.1: Connection refused
telnet: Unable to connect to remote host

Reproduction Steps

$ v -cc clang examples/http_server.v $ examples/http_server

Possible Solution

No response

Additional Information/Context

Very strange. It all works as expected on Linux, so this appears to be FreeBSD specific.

Originally posted by @JalonSolov in https://github.com/vlang/v/issues/16727#issuecomment-1368257371

V version

0.3.2 d19c1ef

Environment details (OS name and version, etc.)

FreeBSD 13.1