Fast, efficient, header-only C Telegram bot API library using FastCGI.
This Telegram library is for anyone who is building serious Telegram bots where performance, scalability and maintainability matter. Whether you're writing a bot to do heavy lifting, servicing thousands of users concurrently through a single webserver, or you just want to write your bot using clean, safe modern C 20, this is your framework.
LibTelegram handles the business of communicating with Telegram for you. It consists of two major components - a listener system, and a sender system. The two are completely decoupled, so you can use both together in a single program, or just one or the other.
The listeners are fully modular - you can choose which listener to use for any program. With all listeners, you set callbacks for those events you're interested in being notified about, and they fire when needed. All listeners automatically handle acknowledgement replies to Telegram, dealing with errors and decoding the incoming json for you.
This listener connects to the Telegram API and requests updates. It uses long polling, leaving a single connection open for as long as possible, so it receives events immediately without delay, and without hammering the Telegram API server with repeated connections. This listener runs single-threaded and calls your callbacks sequentially, in the order of the events received from Telegram. This is a good fit for a standalone bot where persistent operation is required, and a moderately high level of traffic is expected (one or more interactions per minute at all times of day).
** NOTE: FastCGI is disabled in the current version, due to the underlying library becoming outdated vs ASIO. It only with Boost 1.69 and earlier - see darrengarvey/cgi#16. Contributors are welcome to migrate to an alternative fastcgi library. Meanwhile, the polling listener works, and is now the default. **
This listener speaks fastcgi and is designed to be launched by your webserver. It runs multithreaded. You set callbacks, and it asynchronously listens for events from Telegram and calls them when needed. The listener takes care of talking to your webserver and decoding the fastcgi requests. Fastcgi means a single listener can be persistent across many connections, and can handle many connections simultaneously, unlike traditional cgi. Using a webserver based system means you can have hundreds of separate bots running on a single machine, rather than being limited to one per port as with bots that run their own webserver. This listener is a good fit for when you are hosting a large number of bots which spend most of their time idle - the bots will only run when required, and no bandwidth is wasted in establishing polling connections. This is also a good fit for some very traffic-heavy bots that process large amounts of data, such as photographs and file uploads, as it can service many connections from Telegram simultaneously without blocking.
The sender provides a set of convenience functions for sending correctly formatted messages to Telegram - it handles forming the correct data structures, encoding as json, making the web request with SSL, dealing with timeouts, handling and reporting errors, and finally returning to you the results of your query.
All Telegram types are encapsulated in type-safe objects, and elegant error tolerance is provided in a monadic way using std::optional. As well as using the native types, you have the option to access any values decoded from incoming json, and to write the outgoing json yourself, either raw or in convenient json object format.
- Modern, paradigmatic, clean and professional C 20 programming style.
- Header-only - nothing to compile.
- Completely permissive MIT license - no restrictions on what you can do with the library.
- Minimum dependencies - the only external requirement is the ubiquitously available boost_system.
- Modular design - receiving and sending are totally separate, only include the components you use.
- Callback-based architecture means you retain full control of program flow and objects.
- Highly efficient asynchronous, multithreaded architecture allows a single instance to service multiple simultaneous requests at different rates without slowdown or queuing.
- Support for handling any number of bots concurrently from a single instance - instantiate any number of separate standalone listeners and senders.
- Persistent operation - with both the fastcgi and polling listener, the bot keeps running between multiple requests, unlike slow cgi where a new instance has to spawn for each request.
- Persistence means it's also possible to preserve state between calls without resorting to writing to disk or a database!
- Efficient fastcgi protocol - data is transferred from the webserver by a binary format for minimal cpu overhead in repeat encoding and decoding.
No wasteful polling - cgi architecture means the bot is active only when a message from telegram is received, and your callback functions are called instantly rather than after a poll interval.(see above re. FastCGI])- No need to run on its own port, works with your existing fastcgi-capable webserver.
- Each command is optimised and tailored to only send non-default parameters, even when the default is manually specified, to save every little bit of bandwidth.
- Both high and low level callbacks and sending functions available - send messages with a single short command, or craft and decode your own custom json however you like.
- Friendly high-level interface - clean, minimal and easy to use for the most common tasks.
- Low level functionality exposed - you can peek at the headers in Telegram's web requests if you like, or craft raw data to send to telegram by hand if you really want to.
- Detailed real time state monitoring data and lifetime statistics to help you diagnose bottlenecks and squeeze every last drop of performance from your code.
- C 20-compatible compiler with std::optional library support (GCC/G 7 upwards)
- Boost 1.76 upwards - http://www.boost.org
Boost CGI - https://github.com/slowriot/cgi (included, only needed for CGI / FastCGI listener)(see above re. FastCGI])- URDL - https://github.com/chriskohlhoff/urdl (included)
- JSON for Modern C - https://nlohmann.github.io/json/ (included)
- cpp-httplib - https://github.com/yhirose/cpp-httplib (included)
The project is currently in an intermediate stage, so some native convenience functions and types are missing or incomplete - however, almost all functionality is available by reading or specifying the json you require manually through the functions that expose the json types to you.
- Stickers
- Uploading files directly
- Message entities
- Games
- Native functions to export chat links and set chat pictures
All Telegram API types and methods not mentioned above.
The library itself is header-only so there is nothing to build. Simply
#include "libtelegram/libtelegram.h"
and off you go.
A bash script is provided to make it convenient to set the web hook for your bot. This is only required if using FastCGI.
Simply run:
./set_webhook.sh 229029247:AAHfNUvcGS_ttvvulZG3Qi4NmMrjebH8F6w "https://mywebsite.com/fcgi-bin/my_fastcgi_webhook"
(obviously substituting your own token and website).
You can also set the webhook yourself by just sending an appropriate request from your browser, check the telegram bot documentation for the correct syntax.
It's possible to run multiple bots with a single program, and multiple webhooks or polling listeners - see below under "advanced usage".
Consult with the examples for further instruction - this framework aims to be as easy to use as possible, so the examples should be self-documenting.
Simpler examples spend more time explaining simple concepts in comments, while the more advanced examples save the comments for the more advanced features. For that reason, there's a suggested reading order for the examples:
- Minimal echo: examples/echo.cpp
- Echo using raw json access: examples/echo_json.cpp
- Fetch a file from the web and read it back: examples/webfetch.cpp
- Get and process photos sent by the user: examples/getphoto.cpp
- Stateful calculator, using a custom keyboard: examples/calculator.cpp
- Accept locations and update a live location: examples/livelocation.cpp
All examples can be built with CMake, by doing the following:
mkdir build
cd build
cmake ..
make -j$(nproc)
Also included are project files for the Code::Blocks IDE - just open the libtelegram.workspace file in Code::Blocks and you can build all of the examples.
Alternatively, just use the examples as the starting point of your own project, building it with your own build system.
The default examples use polling, so simply execute them in the terminal.
** Note - FastCGI is currently non-functional - see above **
There are a number of fastcgi systems, and it will depend largely on your webserver. Below are some example instructions for how to set up fastcgi on Apache2:
- Install Apache2, obviously.
- Get module
fastcgi
. On debian and noobuntu you canapt-get install libapache2-mod-fastcgi
. - Use the following sort of configuration to enable a directory to execute fastcgi scripts:
ScriptAlias /fcgi-bin/ /var/www/mysite/fcgi-bin/
<Directory /var/www/mysite/fcgi-bin/>
Options ExecCGI
SetHandler fastcgi-script
</Directory>
- Then simply place your executable binaries in this directory, or subdirectories, taking care to give them executable permissions.
- One caveat is that module
fastcgi
does not currently seem to work withmpm_itk_module
running as any user other thanwww-data
on Debian. - Something else to watch out for when testing is that unlike normal cgi, the program persists and keeps running - even after you overwrite the binary. Make sure you either manually kill your bot when you want to update to a new version, or restart the Apache server, or build in functionality for it to exit by itself on command. The webserver will then bring it back up on demand.
- Anything you send to
std::cerr
will appear in/var/log/apache2/error.log
- even if you have vhosts set up with custom logging, the fastcgi module won't send to those. Don't try to output anything tostd::cout
.
All components used in this library should be fully cross-platform compatible; that means at the very least Linux, OS X and Windows. However, it has not been tested anywhere but Linux yet, and the examples and the CMakeLists.txt may require modificaiton to build on other platforms. Let's face it, though, real servers are always going to be Linux.
Have a look at their code for yourself. The primary alternatives to this project
are written in a messy java-like style, and littered with using namespace std
.
The code is full of unnecessary copies and inefficiencies, and fundamental
features such as SSL support are often missing.
- Here is a simple echo example from the leading alternative project: samples/echobot/src/main.cpp
- And here is the same functionality with LibTelegram: examples/echo.cpp
You decide in which style you would prefer to write your next bot!
In order to use CGI, you must set a webhook; a helper script, set_webhook.sh
is provided for your convenience. It takes two arguments - the API key, and the
full https URL of your endpoint. However, if switching back to polling, don't
forget to unset the webhook or you won't receive polling updates! To unset the
webhook, just call the script with no second argument.
There's no requirement to use both the sender and receiver components. If you'd like to make a bot that only sends, then you don't need to use the listener in your program, and it can run standalone rather than through your webserver.
In this case, you may wish to only #include "libtelegram/sender.h"
.
Likewise, if you want to have a bot that only receives but never sends - perhaps for a logging or statistics system - you don't necessarily need to use a sender. If you are using one of the CGI listeners, then no sender is needed. However, the polling listener still needs a sender which it uses to poll Telegram for events.
To send from multiple bots simultaneously, simply create multiple senders, each initiated with the API key of the bot you wish to send as - it's that simple.
To receive from multiple bots with FastCGI or CGI, just use a single listener and symlink your bot to multiple webhook locations on your web server, then distinguish between the bots by examining the URI each callback is called for.
To receive from multiple bots with polling, create a thread for each polling listener to run in. They can each either have their own sender, or can share a single global sender. Senders maintain no global state and create independent connections to the server for each request, thus ensuring lock-free and wait-free thread safety. Calling methods concurrently on the same sender from multiple threads is therefore perfectly safe, should you choose to do that. On the other hand, each sender also has a tiny memory footprint, so creating many senders has very little cost.
If your bot has short-term state you'd like to maintain between requests, you can rely on your fastcgi implementation to keep a single instance alive between requests in order to do so.
However, be aware that you should write anything to disk or DB that it would be a problem to lose - for instance if the webserver undergoes a routine restart, or it decides to cull your fastcgi processes due to a timeout.
If you want to rely on this system for maintaining state in a conversation with a single user, you may need to adjust your fastcgi server's settings to only allow a single concurrent instance - otherwise you may find your webserver runs two or more instances of your bot, and the same user may be serviced by two different instances on subsequent requests; the second instance would of course have no memory of servicing that user before, even if the instance was already alive at the time.
If you're running the polling listener, your program will run as long as there
are no fatal errors with the listener, and you can get away with saving state
after listener.run()
has completed. However, it would be wise to also save
your state periodically in case your bot crashes or is killed by something.
Please don't do this, for the sake of your users. The costs of ssl are minimal,
and passing your users' communications in the open is simply barbaric. But this
is your bot, so if you really feel you must turn off SSL, then define
LIBTELEGRAM_DISABLE_SSL_NO_REALLY_I_MEAN_IT_AND_I_KNOW_WHAT_IM_DOING
before you include LibTelegram's headers. But don't say we didn't warn you.
The polling listener (and possibly other future components) sets a signal handler to catch ctrl-c and shut down neatly, after the next poll. This is currently the only elegant way of shutting down listener::run() for the polling listener; it is assumed that in a CGI or FCGI program, the webserver will terminate its children immediately whenever it wants, so there is no attempt to handle signals in other listeners. The listener only sets the signal handler at the beginning of run() and unsets it at the end.
You can also manually set or release this signal handler at any time using the
telegram::listener::poll::set_signal_handler()
and unset_signal_handler()
functions, and check its current state with is_signal_handler_set()
.
However, if you want to set your own signal handlers, and don't want this to
interrupt your global state, then define TELEGRAM_NO_SIGNAL_HANDLER
in your
program. In this case, be aware that run() will continue indefinitely until you
tell it to stop (from another thread); to do this you can call the
telegram::listener::poll::stop()
function. You can also call stop_all()
to
queue a global halt for all poll listeners that exist in your program (which is
what the signal handler does).
Windows users: The default signal handler requires sigaction
and will not
compile on Windows; as a result, TELEGRAM_NO_SIGNAL_HANDLER is defined by
default for Windows builds.
You can exclude individual modular components you're not using, which may remove compile time dependencies on certain libraries. Specific parts can be disabled at compile time with:
#define TELEGRAM_NO_LISTENER_FCGI
#define TELEGRAM_NO_LISTENER_POLL
This project is an early work in progress, and contributions are always welcome. The following areas especially require attention at present:
- Tests - if you'd like to write tests for any components and set up test coverage for them, it would be very helpful.
- Additional convenience functions - any parts of the Telegram API that aren't covered by their own sending functions or callbacks can be added.
Contributors are asked to adhere to the VoxelStorm Style Guide - code not adhering to the style guide may be accepted, but will be converted to match the style guide.
This is not an exhaustive list; if you are building a project using this library, please contribute your project to this list.