Facil.io: A micro-framework for C web applications

by anderspitman • 7 years ago

119 42 7 years ago

scegit2• 7 years ago

Pretty cool! but it doesn't support HTTP2 which is a bit disappointing

lukeqsee• 7 years ago

Very few folks don’t run some reverse proxy in front of their application servers (or a whole string of them to load balance), so lack of http/2 on an application server is rarely a dealbreaker for a web framework. I’ll take a solid framework with only h/1.1 over an iffy one with h/2 any day of the week.

loeg9• 7 years ago

The FreeBSD project has been dealing with HTTP2 interoperability hell over the last month. (Curl 7.62 may or may not talk to Phabricator if HTTP2 is enabled; it seems to depend on some local conditions and the phase of the moon whether it works or not. HTTP1.1 just works.) Given that experience with HTTP2, I'm quite happy to stick to HTTP1.1 for now.

scegit2• 7 years ago

HTTP1.1 is expected to just work it has been here for a very long time. It takes time to understand, support and adopt new technology. I'm not sure if the blame should be on HTTP2 in this case.

polotics• 7 years ago

The example has functions' string parameters and separate parameters for the string's length. Is this really necessary what happens if the declared length is more than the actual length?

tom_mellior3• 7 years ago

Knowing nothing about this, I would guess that this is to allow sending binary data that contains embedded null bytes?

Though I guess it would be good to have nicer wrappers for the common case where you do only send text.

loeg9• 7 years ago

Probably they use length-prefixed strings internally (which is probably a requirement for taking a C web framework seriously), but allow C nul-terminated strings at the API boundary for convenience of use.

pantalaimon• 7 years ago

No it is not necessary. If no length is given, the string is assumed to be NULL-terminated.

It's a bit unfortunate that they use two code paths for that which are basically just copy & paste.

https://github.com/boazsegev/facil.io/blob/master/lib/facil/...

bowild2• 7 years ago

This approach moves an `if` statement out of the loop, so there's no need to repeat it.

bowild2• 7 years ago

Actually, it's to allow for non-NUL terminated strings.

A good example is when using a String supplied by a scripting language such as Ruby or JavaScript.

Normally, you might need to copy the string to add a NUL byte, or request that the scripting engine provide a NUL terminated string (which might result in the same action performed).

Using this optional approach, a potential copy of the data can be avoided.

tom_mellior3• 7 years ago

I like the use of keyword argument syntax to specify optional arguments. I haven't seen anything like this before in C or C++, and now I would wish for it to become standard.

It's simulated cleverly using a macro that wraps variable arguments in a struct literal: See the definition of http_set_cookie in https://github.com/boazsegev/facil.io/blob/master/lib/facil/...

saagarjha• 7 years ago

I haven't seen anything like this before in C or C++, and now I would wish for it to become standard.

I've seen people try to simulate this in C++ with tags.

mafm• 7 years ago

I think this trick comes from Ben Klemens' "21st Century C".

If you like this, the book has some other tricks in it you might also like. (One example: you can define generic functions in C11).

jolmg2• 7 years ago

That's really cool, but don't unassigned fields have undefined values? How would the function know what was set? I imagine that in practice they're almost always 0, but that's not guaranteed, right? You'd need to set the struct as {0}, to have everything be 0. Is this a feature of a new(ish) standard?

hawski• 7 years ago

It is guaranteed per C99. It's a compound initializer.

Also if you set a struct with {1} the first field is set to 1 and the rest to 0. AFAIR it was always like that (at least since standardized C, but probably earlier).

loeg9• 7 years ago

C99 specifies that struct literals with absent fields are assigned a zero value. (Notably unspecified is the value of any implicit padding bytes in the struct.)

drudru112• 7 years ago

Yes, this exactly - as soon as I saw that, I was like “how did he pull that off?”.

Of course, the first comment on HN has the answer.

I am going to try this trick out today.

drudru112• 7 years ago

The exact line is here:

https://github.com/boazsegev/facil.io/blob/master/lib/facil/...

I've seen other hacks to do this, but this is the cleanest that I've seen. He uses the same name for the macro and the function.

The only thing I would change is the way the struct is passed. I would pass by reference.

I really like this.

beefhash• 7 years ago

If you want to do web applications in C, you may also be interested in the kcgi[1] and ksql[2]. kcgi takes a more traditional approach by using CGI rather than embedding a static file server in the final binary.

[1] https://kristaps.bsd.lv/kcgi/

[2] https://kristaps.bsd.lv/ksql/

jondubois• 7 years ago

CGI adds simplicity and interoperability but takes away control and performance. FastCGI suffers from the same fundamental issue as well but to a lesser extent.

EdSchouten• 7 years ago

There is also Ulfius: https://github.com/babelouest/ulfius/

striking• 7 years ago

And https://kore.io

bitconion• 7 years ago

and worth to mention libh2o in h2o https://github.com/h2o/h2o

pvtmert• 7 years ago

only thing i did not understand is port being string/char*

other than that, clever api

loeg9• 7 years ago

Probably for getaddrinfo():

http://beej.us/guide/bgnet/html/single/bgnet.html#getaddrinf...

ape4• 7 years ago

I see you can use a name (rather than a number) for the port like "ftp" or "http".

badrabbit2• 7 years ago

Honset curiosity: is there a reason Golang or rust couldn't perform same or better? I get why as a programmer I would use C, but what benefit does it have for the users of the app?

jolmg2• 7 years ago

The users of the framework, the programmers of a webapp? I imagine that there's few frameworks for web development in C, so having options is good.

It's also probably way more portable than Golang or rust. That's also useful for the users of the webapp if they want to host it. I mean, it probably takes less effort to get it to compile and run on a system like SCO, compared to doing that with Golang or rust. You'd think legacy systems don't matter, but I imagine that there's still many businesses that heavily depend on such legacy systems because the costs to safely migrate are big.

Making it in C is probably also going to result in far smaller compiled code, since I imagine C requires a far smaller RTS than Golang or rust. That might make a difference for people that for some reason want to run a webapp on a small device.

I may be wrong though, since I don't have any experience with Golang or rust.

badrabbit2• 7 years ago

I meant the users of the webapp.

loeg9• 7 years ago

Sure. Go is a garbage-collected language with an immature and unsophisticated compiler toolchain, compared to the big free C compilers like GCC or Clang. The compiler limitations are solvable but GC is a core part of the language. Interoperability with C libraries is possible but not great.

Rust is built on LLVM, much like Clang. C integration is relatively easy. I think this is more viable than Go. Unfortunately, Rust is only portable to platforms LLVM supports. While that covers the majority of commonly used hardware today, it is not comprehensive. Also, Rust compile times are pretty bad, especially for optimized builds.

At the end of the day the author gets to write code in a language of their choice to scratch their own itch. End users don't care at all what language a web framework is written in.

oritsnile• 7 years ago

As far as I know rust isn't garbage collected

burntsushi2• 7 years ago

Rust is no more a garbage collected language than C is. Just as one can implement reference counting in C, one can do the same in Rust, but it is not forced on you just like it is not forced on you in C.

IloveHN84• 7 years ago

Where's TLS? And http/2?

doctorpangloss• 7 years ago

People who serve SSL out of their Java servlets by installing certificates to random Tomcat and Linux system directories are already a little crazy.

You'd have to be a real maniac to serve TLS in someone's random C framework.

Facil.io: A micro-framework for C web applications

Related Stories