Core c99 package for AWS SDK for C. Includes cross-platform primitives, configuration, data structures, and error handling.
This library is licensed under the Apache 2.0 License.
aws-c-common uses CMake for setting up build environments. This library has no non-kernel dependencies so the build is quite simple.
For example:
git clone git@github.com:awslabs/aws-c-common.git aws-c-common
mkdir aws-c-common-build
cd aws-c-common-build
cmake ../aws-c-common
make -j 12
make test
sudo make install
Keep in mind that CMake supports multiple build systems, so for each platform you can pass your own build system
as the -G
option. For example:
cmake -GNinja ../aws-c-common
ninja build
ninja test
sudo ninja install
Or on windows,
cmake -G "Visual Studio 14 2015 Win64" ../aws-c-common
msbuild.exe ALL_BUILD.vcproj
Every API has a specific set of styles and conventions. We'll outline them here. These conventions are followed in every library in the AWS C SDK ecosystem.
Every function that returns an int
type, returns AWS_OP_SUCCESS
( 0 ) or AWS_OP_ERR
(-1) on failure. To retrieve
the error code, use the function aws_last_error()
. Each error code also has a corresponding error string that can be
accessed via the aws_error_str()
function.
In addition, you can install both a global and a thread local error handler by using the aws_set_global_error_handler_fn()
and aws_set_thread_local_error_handler_fn()
functions.
All error functions are in the include/aws/common/error.h
header file.
Any function that allocates and initializes an object will be suffixed with new
. Similarly, these objects will always
have a corresponding function with a destroy
suffix. The new
functions will return the allocated object
on success and NULL
on failure. To respond to the error, call aws_last_error()
. If several new
or destroy
functions are available, the variants should be named like new_x
or destroy_x
(e.g. new_copy
or destroy_secure
).
Any function that initializes an existing object will be suffixed with init
. These objects will have a corresponding
clean_up
function if necessary. In these cases, you are responsible for making the decisions for how your object is
allocated. The init
functions return AWS_OP_SUCCESS
( 0 ) or AWS_OP_ERR
(-1) on failure. If several init
or
clean_up
functions are available, they should be named like init_x
or clean_up_x
(e.g. init_static
or
clean_up_secure
).
If you are contributing to this code-base, first off, THANK YOU!. There are a few things to keep in mind to minimize the pull request turn around time.
These "guidelines" are followed in every library in the AWS C SDK ecosystem.
- All APIs that need to be able to allocate memory, must take an instance of
aws_allocator
and use that. Nomalloc()
orfree()
calls should be made directly. - If an API does not allocate the memory, it does not free it. All allocations and deallocations should take place at the same level. For example, if a user allocates memory, the user is responsible for freeing it. There will inevitably be a few exceptions to this rule, but they will need significant justification to make it through the code-review.
- All functions that allocate memory must raise an
AWS_ERROR_OOM
error code upon allocation failures. If it is anew()
function it should return NULL. If it is aninit()
function, it should returnAWS_OP_ERR
.
- Occasionally a thread is necessary. In those cases, prefer for memory not to be shared between threads. If memory must cross a thread barrier it should be a complete ownership hand-off. Bias towards, "if I need a mutex, I'm doing it wrong".
- Do not sleep or block .... ever .... under any circumstances, in non-test-code.
- Do not expose blocking APIs.
- For APIs returning an
int
error code. The only acceptable return types areAWS_OP_SUCCESS
andAWS_OP_ERR
. Before returning control to the caller, if you have an error to raise, use theaws_raise_error()
function. - For APIs returning an allocated instance of an object, return the memory on success, and
NULL
on failure. Before returning control to the caller, if you have an error to raise, use theaws_raise_error()
function.
The error handling infrastructure is designed to support multiple libraries. For this to work, AWS maintained libraries have pre-slotted error codes for each library. The currently allocated error ranges are:
Range | Library Name |
---|---|
[0x0000, 0x0400) | aws-c-common |
[0x0400, 0x0800) | aws-c-io |
[0x0800, 0x0C00) | aws-c-http |
[0x0C00, 0x1000) | aws-c-compression |
[0x1000, 0x1400) | aws-c-eventstream |
[0x1400, 0x1800) | aws-c-mqtt |
[0x1800, 0x1C00) | (reserved for future project) |
[0x1C00, 0x2000) | (reserved for future project) |
[0x2000, 0x2400) | (reserved for future project) |
[0x2400, 0x2800) | (reserved for future project) |
[0x2800, 0x2C00) | (reserved for future project) |
[0x2C00, 0x3000) | (reserved for future project) |
[0x3000, 0x3400) | (reserved for future project) |
[0x3400, 0x3800) | (reserved for future project) |
[0x3800, 0x3C00) | (reserved for future project) |
[0x3C00, 0x4000) | (reserved for future project) |
Each library should begin its error codes at the beginning of its range and follow in sequence (don't skip codes). Upon adding an AWS maintained library, an error code range must be approved and added to the above table.
We have a high bar for test coverage, and PRs fixing bugs or introducing new functionality need to have tests before they will be accepted. A couple of tips:
We provide a test harness for writing unit tests. This includes an allocator that will fail your test if you have any
memory leaks, as well as some ASSERT
macros. To write a test:
- Create a *.c test file in the tests directory of the project.
- Implement one or more tests with the signature
int test_case_name(struct aws_allocator *, void *ctx)
- Use the
AWS_TEST_CASE
macro to declare the test. - Include your test in the
tests/main.c
file. - Include yur test in the
tests/CMakeLists.txt
file.
- No Tabs.
- Indent is 4 spaces.
- K & R style for braces.
- Space after if, before the
(
. else
andelse if
stay on the same line as the closing brace.
Example:
if (condition) {
do_something();
} else {
do_something_else();
}
- Avoid C99 features in header files. For some types such as bool, uint32_t etc..., these are defined if not available for the language
standard being used in
aws/common/common.h
, so feel free to use them. - For C++ compatibility, don't put const members in structs.
- Avoid C++ style comments e.g.
//
. - All public API functions need C++ guards and Windows dll semantics.
- Use Unix line endings.
- Where implementation hiding is desired for either ABI or runtime polymorphism reasons, use the
void *impl
pattern. v-tables should be the last member in the struct. - For #ifdef, put a # as the first character on the line and then indent the compilation branches.
Example:
#ifdef FOO
do_something();
# ifdef BAR
do_something_else();
# endif
#endif
- For all error code names with the exception of aws-c-common, use
AWS_ERROR_<lib name>_<error name>
. - All error strings should be written using correct English grammar.
- SNAKE_UPPER_CASE constants, macros, and enum members.
- snake_lower_case everything else.
static
(local file scope) variables that are notconst
are prefixed bys_
and lower snake case.- Global variables not prefixed as
const
are prefixed byg_
and lower snake case. - Thread local variables are prefixed as
tl_
and lower snake case. - Macros and
const
variables are upper snake case. - For constants, prefer anonymous enums.
- Don't typedef structs. It breaks forward declaration ability.
- Don't typedef enums. It breaks forward declaration ability.
- typedef function definitions for use as function pointers as values and suffixed with _fn.
Example:
typedef int(fn_name_fn)(void *);
Not:
typedef int(*fn_name_fn)(void *);
- Every source and header file must have a copyright header (The standard AWS one for apache 2).
- Use standard include guards (e.g. #IFNDEF HEADER_NAME #define HEADER_NAME etc...).
- Include order should be: the header for the translation unit for the .c file newline header files in a directory in alphabetical order newline header files not in a directory (system and stdlib headers)
- Platform specifics should be handled in c files and partitioned by directory.
- Do not use
extern inline
. It's too unpredictable between compiler versions and language standards. - Namespace all definitions in header files with
aws_<libname>?_<api>_<what it does>
. Lib name is not always required if a conflict is not likely and it provides better ergonomics. init
,clean_up
,new
,destroy
are suffixed to the function names for their object.
Example:
AWS_COMMON_API
int aws_module_init(aws_module_t *module); AWS_COMMON_API void aws_module_clean_up(aws_module_t *module); AWS_COMMON_API aws_module_t *aws_module_new(aws_allocator_t *allocator); AWS_COMMON_API void aws_module_destroy(aws_module_t *module);
- Avoid c-strings, and don't write code that depends on
NULL
terminators. Exposestruct aws_byte_buf
APIs and let the user figure it out. - There is only one valid character encoding-- UTF-8. Try not to ever need to care about character encodings, but where you do, the working assumption should always be UTF-8 unless it's something we don't get a choice in (e.g. a protocol explicitly mandates a character set).
- If you are adding/using a compiler specific keyword, macro, or intrinsic, hide it behind a platform independent macro definition. This mainly applies to header files. Obviously, if you are writing a file that will only be built on a certain platform, you have more liberty on this.