Somehow it still works because the lower levels can deal with
inconsistent/broken state, but this lets us avoid that broken state in
the first place. If a friend connection doesn't exist, we don't add it to
our group.
The void pointer here only adds opportunity to introduce bugs and doesn't
actually make things more layered. It's just the code lying about being
layered while it's actually spaghetti.
The enumerators won't change, but the type name will change in 0.3.0.
Reasoning:
- Type names in toxcore start with an uppercase letter and either have at
least one lowercase letter in them, or are less than 4 characters long.
- Constants consist of 4 or more uppercase letters or underscores.
By these rules, "DHT" is a type name, but "TOX_USER_STATUS" is a
constant. We provide Tox_User_Status as an alternative for now, and will
switch to that in 0.3.0, removing the UPPER_CASE versions.
Presumably the uses of `rand()` were fine because they were not used in
security-sensitive places, but having to think about whether a crappy RNG
is acceptable in each situation requires effort that could better be
spent elsewhere.
Also, this means that once we have a custom deterministic RNG for
testing, that RNG is used everywhere, so all the code is deterministic.
It also allowed us to delete a system-specific function that wasn't used
anywhere except in a call to `srand()`.
This allows Tox to contain additional data on top of Messenger, making
Messenger not necessarily the most top-level object. E.g. groups are
built on Messenger and currently awkwardly void-pointered into it to
pretend there is no cyclic dependency.
All for-loops in toxcore are of the form
for (<for-init>; <for-cond>; <for-next>) { <body> }
`for-init` can be a variable declaration (like `int i = 0`), an
assignment (like `i = 0`), or empty.
`for-cond` can be any expression.
`for-next` can be an assignment or a single increment/decrement
expression (like `++i` or `--i`).
No other forms are allowed, so e.g. comma expressions in any of these are
not allowed (so no `for (i = 0, j = n; ...; ++i, --j)`).
* test names in conference_test
* raise error on attempt to invite friend to group before we are connected
* revise handling of temporary invited connections
We are now careful not to prematurely delete a connection to a peer
established during the invitation process; namely, before we have sufficient
other connections and have confirmed that we have an alternative route to the
peer.
* process out-of-order messages from a peer
* don't reset names when handling a Peer Response
`UID` sounds like `User ID`. While it is a Unique ID, the property of an
"identifier" is generally that it identifies a unique thing, so the 'U'
is redundant, and `GUID` as a globally unique id (which is likely also
true for these IDs) has a specific meaning and syntax, so we're not using
that. So, we just say conference `id`.
Also, added some #defines to make symbols visible that are in BSD but not
in UNIX. Solaris needs these, since it's fairly strict with its symbol
visibility in system headers.
256 bytes including NUL byte is confusing and makes for really annoying
bindings to other languages that don't account for NUL bytes in their
string length. We pass C strings, not byte arrays, for hostnames, so 255
makes more sense here.
By changing numchats from uint32_t to uint16_t. This is done in PGC. This
PR is making that change in master to reduce the diff in the PGC branch.
Also:
* Inverted groupnumber_not_valid and renamed to is_groupnumber_valid.
* Renamed realloc_groupchats to realloc_conferences and made it return bool.
* Added setup_conference function that currently just zeroes the
conference structure but later will initialise more values.
* Made some `i` iterator variables local to the for-loop using
for-init-decl. This is also done in PGC.
* Moved PAIR to toxav, where it's used (but really this should die).
* Replace most MIN calls with typed `min_*` calls. Didn't replace the
ones where the desired semantics are unclear. Moved the MIN macro to
the one place where it's still used.
* Avoid assignments in `while` loops. Instead, factored out the loop body
into a separate `bool`-returning function.
* Use named types for callbacks (`_cb` types).
* Avoid assignments in `if` conditions.
* Removed `MAKE_REALLOC` and expanded its two calls. We can't have
templates in C, and this fake templating is ugly and hard to analyse
and debug (it expands on a single line).
* Moved epoll system include to the .c file, out of the .h file.
* Avoid assignments in expressions (`a = b = c;`).
* Avoid multiple declarators per struct member declaration.
* Fix naming inconsistencies.
* Replace `net_to_host` macro with function.
* Removed `ARRAY_SIZE` and use NULL markers for end of array, instead.
The alternative is + size, but for these arrays, NULL markers made
sense, since they are arrays of non-null pointers.
* Made `INDEX_OF_PK` a self-contained macro, not dependent upon the
naming inside its call site. This is a minor change but makes the code
more local and reviews easier.
* No nested structs.
* Use only named function types ending in `_cb` for callbacks.
* Replaced two macros with functions.
* `++i` instead of `i++`.
* struct member names start with lowercase letters.
* It takes a bit of work to support `/**/` comments in preprocessor
macros, so I've decided not to support these. If a macro is complex
enough to need comments inside it, it's too complex. `//` comments are
allowed at the end of macro definitions.
* Callback typedefs must name their parameters.
* Enums must by typedef'd.
* Comments at end of `#define` must be `//` comments.
* Typedef structs must not be anonymous.
* `;` at the end of a `#define` is invalid.
* Callback typedefs must list their parameter names.
* No nested structs.
* No inline use of function pointer types. Only typedef'd callback types
are allowed.
* Enum types are spelled in Camelsnake_Case.
* The argument to `#error` must be a string literal.
It turns out, `unix_time` is also monotonic, and is used as such, so I've
renamed the new functions to `mono_time_*`.
2018-07-08:
```
00:01 <@irungentoo> the idea used to be that the unix_time() function
could go backward in time but I think I might have started using it like
if it could not after I changed it so that it would never go back in time
```
Also:
* compound statements (blocks, e.g. in if/else) must be non-empty.
Comments don't count.
* `=` is not allowed in subexpressions. We treat it as statement-only.
* `++i` is preferred over `i++` in statements.
* `Type_Names` are camelsnake-case.
Also:
* `#define` must be scoped. If it's used outside a scope, it must be
defined outside that scope (global `#define` does not need a matching
`#undef`).
This is now a style rule: you can only use typedef'd function types.
Previous rules now applied in `onion_*.c`:
* `struct`s must have a name (typedef of unnamed struct is not allowed).
* `++i` for increment-stmt, not `i++`, e.g. in loops.
* Only a single declarator per struct member declaration.
* Type_Names vs. variable_names.
Also:
* No inner structs.
* One declarator per member declaration.
* Function names are snake_case.
* Names ending in `_cb` are function types.
* `++i` is preferred over `i++`.
Rules:
1. Constants are uppercase names: THE_CONSTANT.
2. SUE[1] types start with an uppercase letter and have at least one
lowercase letter in it: The_Type, THE_Type.
3. Function types end in "_cb": tox_friend_connection_cb.
4. Variable and function names are all lowercase: the_function.
This makes it easier for humans reading the code to determine what an
identifier means. I'm not convinced by the enum type name change, but I
don't know a better rule. Currently, a lot of enum types are spelled like
constants, which is confusing.
[1] struct/union/enum
iphydf