* Module rewrite
- WinSock version changed from 1.1 to 2.2.
- Properly check for WinSock initialization on OnAmxxAttach/Detach.
- Now natives will not be added if we can't start up WinSock.
- socket_open() is now IP version agnostic (both IPv4 and IPv6 are
supported).
- Error reporting has been changed on socket_open(), a new parameter
called _libc_errors has been added, and, if enabled, libc errors will be
returned instead of the previous made-up errors.
- socket_close() now returns a value on success/failure.
- Added non-blocking sockets at socket_open_nb().
- Added socket_is_writable() to check if a socket is ready for write.
- Added socket_is_readable() as an alias to socket_change().
- Code rewritten to be more readable, it should be self-explaining now.
* Update docs and fix AMBuild
Updated documentation following the guidelines
* Fixs for the module
- Fixed the backwards compatibility with the return codes
- Merged socket_connect and socket_connect_nb
- Added a 5th parameter to socket_open that takes bit flags to enable
the new features (libc errors & nonblocking sockets)
- Fixed an error on socket_send2 that caused the buffet not to start
from the beginning if multiple calls were made
- Updated docs
- [docs] Prefixed error codes with SOCK_
- [docs] Added the new flags SOCK_NON_BLOCKING and SOCK_LIBC_ERRORS
- [docs] Added a new stock called SOCK_ERROR_EINPROGRESS(error) to be
used when checking if a newly created nonblocking socket is connecting
* Fixes for the docs
Fix some typos, shorten lines, document SOCK_ERROR_EINPROGRESS
* Document magic numbers
No more magic
* Revert "Document magic numbers"
This reverts commit 0f233292063400ea7fdbcd5e5d5cd6e54f8cd71c.
* More docs fixes
* Fix broken socket_send2()
* Add error checking in send2()
* Remove stock
It should not be needed because nb sockets should always be checked or
writability
* Fix some identations
* Fix return codes doc
* Fix socket_recv() regression
strncopy will stop on a null byte, that makes the function unusable to
receive binary data
* More docs typo fixes
* Rename client_disconnected to client_disconnecting
* Add client_disconnected as post forward
* Fix client not properly disconnected internally
Introduced in #264.
Edict is reset once SV_DropClient is called, so that second check would be always false.
* Reflect changes on the concerned plugins
* Revert renaming, let's add only client_remove as post forward
* Revert "Extend "read_argv" native"
This reverts commit aaa2934595.
This broke binary compatibility with either older, already compiled
plugins, or newly compiled plugins if you forgot to update .inc headers.
This happened because read_argv used to receive maxlen by value, and
after
this commit it receives it by reference. This causes read_argv either
to fail, or worse, to buffer overflow, resulting in a security vulnerability.
Newly introduced functionality for read_argv should be added with a new,
separate native.
* Add "read_argv_int" & "read_argv_float" natives
The api says that the user_silentkill() "Kills a user without a message.", that means it is just user_kill() without sending deathmsg.
But user_silentkill() automatically kills player without affecting the client's score (and that is not mentioned in api), so i'll add a second parameter to user_silnetkill() to let user to decide if he want to kill a player without or not affecting the table.
New parameter has the default value 1, so it won't break the compatibility.
- The logic around CS_OnBuy forward has been simplified. Since there is no way to have a consistent way to hook/block for all items, the new logic is to have as less as possible code, especially in blocking mode where we want to avoid to do extra stuffs (e.g blocking sound, event, etc).
* All guns + shield -> CanBuyThis()
* Nvgs and Fefuser only -> CanPlayerBuy()
* The others items -> GiveNamedItem() + AddAccount()
* Ammos -> -> BuyGunAmmo() + GiveNamedItem() + AddAccount()
- Fixed missing buyzone check when alias from console are used (CS_OnBUy* were incorrectly fired).
- Fixed an infinite loop when buying of ammos are blocked. Sorted by hooking BuyGunAmmo().
- Fixed blocking mode for some items. Some game behaviors were not blocked (e.g. weapon drop).
- Fixed forwards being triggered even though errors were found. Detours are now a destroyed and associated variables resetted when necessary. Toggling forwards state is now based on detours state.
- Moved things in its own functions (game functions to execute, class members retrieval)
- Renamed CommandAliases -> ItemInfos (more generic)
This is actually not possible to have a proper system to allow this.
This is two majors problems:
- The item price text can't be changed, whatever old and VGUI menu (it's either harcoded in config file or in client binary)
- Once you open VGUI menu, to know if user has enough money to select an item, client relies on the current HUD money value, this means, since we can't directly changed price, before opening the menu, money needs to be somehow faked and restored once closed. It's awful.
Overall it can't work properly, and at the end current forward doesn't make sense.
If an author wanted to have its own prices, the only only way would to force players to use old menu, then overwriting the whole buy menu, so you would be able to display what you want exactly.
Related to #266. Looks like I've changed the prefix afterward at some point for some reason, but engine is retarded and with `plugin.something.cfg` it will understand extension is `something` instead of `cfg` ; and will fail to execute as it's an invalid extension (only `cfg` and `rc` are allowed).
- KVD_Wrapper compatibility with Ham by making |&kvdw == &kvdw.kvd|
- Replaces g_kvd_hook with a generic g_kvd_ext to support raw KVD from Ham
- Adds create_kvd() and free_kvd() so plugins have something to feed Ham with
- The "ALL" fake pathID is replaced by what does SM, having a public var NULL_STRING which will acts as NULL when needed.
To make compiler accepting public array, this patch was needed: https://hg.alliedmods.net/sourcemod-central/rev/b12f329def09
- The offset thing in read_dir: considering that's something very specific to this native and that implementation in CDirectory doesn't make sense because of the offset compatibility for windows, all code is now in the native.
