Only handle self-presence for rooms we are actively managing.
Earlier we blindly react to any self-presence
stanza by calling 'ui_room_join'.
The join could have been initiated by another client or an external
Gateway. In the case of #731 this was Slack.
I remember that more people reported this kind of bug in our MUC.
We then would open a window and the nick would be (null) because we
don't have any nickname for that room set.
So only react on self-presence if we manage the MUC.
Other XMPP clients do the same.
Fix https://github.com/profanity-im/profanity/issues/731
When an OMEMO bundle is received with an empty <prekeys/> element,
`prekeys_list` remains empty. In omemo_start_device_session(),
performing a modulo operation on the length of an empty list
(0) triggered a SIGFPE (Floating point exception).
torsocks likely affected the network environment where such empty
or incomplete bundles could be intercepted.
Fix https://github.com/profanity-im/profanity/issues/1997
Should help us find the following bugs early.
ASan:
* Out-of-bounds accesses
* Use-after-free
* Memory leaks
* Double free / Invalid free
UBSan:
* Undefined Behavior
* Signed integer overflow
* Null pointer dereference
* Pointer misalignment
* Division by zero
* Bit-shifting out of bounds
This needs works during runtime. But let's add it here so that at least
basic --version run is test and as a reminder for developer to add it.
I will also adjust documentation later on.
Simplify how field values are updated and ensure memory is correctly
managed.
The previous implementation relied on manual manipulation of the
GSList internal `data` member and only handled cases where a field
had zero or one existing value.
The function now correctly handles fields that may
already contain multiple values by using 'g_slist_free_full' to
clear the entire list and its contents before setting the new
value.
Adding an explicit NULL check for 'input' at the start of
cmd_ac_complete to prevent a potential crash and resolve
a GCC static analyzer warning.
Quirk Explanation:
The analyzer found a "deref-before-check" warning for:
`if ((strncmp(input, "/", 1) == 0) && (!strchr(input, ' ')))`
The analyzer interpreted 'strchr(input, ' ')' as a point where the
validity of 'input' is being questioned (a "check"), even though it is
actually checking the return value. This is due to GCC's internal model
of certain standard string functions having a 'nonnull' attribute or
being categorized as pointer "interrogations" in its state machine.
Because 'strncmp' dereferences 'input' earlier in the same line, the
analyzer saw a logical contradiction: "You treat it as safe in strncmp,
but then you call a function (strchr) that 'requires/checks' it to be
safe, implying you weren't sure."
Even though we are using with GNU extensions I will add this flag
since I prefer the explicit writing style that we have to use wich
this flag enabled.
Let's build with -fanalyzer in debug mode.
These commits are related to issues found by gccs static analyzer.
Also bulid with -fstack-protector-strong compiler flag.
This flag adds a canary to stack frames, causing the program to terminate immediately if stack corruption is detected.
This helps identify memory safety bugs earlier during development by
turning silent corruption into an immediate crash.
* _rotate_log_file(): guard against NULL mainlogfile before g_strdup/strlen
* log_stderr_init(): move dup2 after malloc checks to avoid fd leak on
allocation failure.
Also remove explicit close(STDERR_FILENO) before dup2
since dup2 closes the target atomically. Close the fd returned by
dup2 immediately.
We were sequentially checking for errors from 'curl_easy_perform',
'ftell', and 'fclose'. Each time overwriting the last one, resulting in
a leak. This commit ensures that 'err' is only set if it is currently
NULL, preserving the first and most specific error encountered.
`userdata` was not freed if a handler could not be registered.
This occurs when the 'id' parameter is NULL. In such cases, the
handler cannot be stored in the 'id_handlers' table, but ownership
of 'userdata' has already been transferred to this function. This
commit ensures 'free_func' is called if 'id' is NULL since we always
need to have an ID.
recp elements were accessed without verifying the success of
_ox_key_lookup().
Also fix several memory leaks.
gpgme_data_new() followed 'gpgme_data_new_from_mem' was redundant.
Usually we use `/help command`. Additionally there is a shortcut
`/comamnd?`. 54fea4afcf tried to introduce
to have `/command help`. Which doesn't really work since most commands
get this as parameters. So it never worked. Additionally it introduced
at segfault because it tried to set *question_mark to NULL even we can
also get there without the `?` but by having the `help` as parameter.
Modernize the /export command implementation by adopting GLib-based
patterns for string manipulation and file handling.
* Replace _writecsv with _append_csv_escaped to efficiently handle CSV
quote doubling directly within a GString buffer
* Refactor cmd_export to build the CSV content in memory using GString,
minimizing system calls
* Use g_file_set_contents for atomic writing
* Cache strlen() result
* Check for malloc() failure to prevent null pointer dereference
* Ensure the output buffer is null-terminated for safe use with cons_show()
* Simplify the quote escaping logic for better readability.
Make get_random_string() return a gchar*.
This is part of the effort to migrate string declarations and
allocations from char to gchar, replacing standard C allocators
(malloc/calloc/strdup) with their glib equivalents
(g_malloc/g_new0/g_strdup).
The primary goal is to prevent mismatched allocator bugs. While char and
gchar are binary compatible, mixing their respective deallocators is
dangerous:
* Freeing malloc'd memory with g_free() (or vice versa) can lead to
memory corruption, double-frees, or crashes.
* This is seems to be the case especiall on non Linux platforms or when
using hardened memory allocators where the GLib slice allocator or
wrappers may differ from the system heap.
By standardizing on gchar, we ensure that our automatic cleanup macros
(like auto_gchar) always invoke the correct deallocator (g_free).
Migrate structure allocations from calloc to glibs g_new0 macro to
improve type safety and memory robustness.
* Type Safety: The macro takes the type name directly, ensuring the
allocated size always matches the pointer type.
* Static Analysis: It guarantees a non-NULL return by aborting on
failure, which silences -fanalyzer warnings regarding potential NULL
pointer dereferences.
* Readability: Removes redundant sizeof() calls and is the glib way
