<!--
Please provide as much information as possible about what your PR aims
to do.
PRs with no description will most likely be closed until more
information is provided.
If you're planing on changing fundamental behaviour or add big new
features, please open a GitHub Issue first before starting to work on
it.
If it's not something big and you still want to contact us about it,
feel free to do so !
-->
### Problem description
<!-- Describe the bug that you fixed/feature request that you
implemented, or link to an existing issue describing it -->
Fixed possible bug of `EventManager::unsubscribe`
`std::map` only allows unique key, but the same token can subscribe to
multiple events.
1a2a926b77/lib/libimhex/include/hex/api/event.hpp (L104-L107)
If the previous token has already subscribed to an event, then when
subscribing again, `getTokenStore().insert` will not do anything
(Because its type is `std::map`)
1a2a926b77/lib/libimhex/include/hex/api/event.hpp (L122-L134)
At this point in `unsubscribe`, the `iter` may not be able to find the
correct event and erase it
### Implementation description
<!-- Explain what you did to correct the problem -->
Change `tokenStore` to `std::multimap` instead of `std::map`, which
cannot unsubscribe multiple events correctly
### Screenshots
<!-- If your change is visual, take a screenshot showing it. Ideally,
make before/after sceenshots -->
### Additional things
<!-- Anything else you would like to say -->
<!--
Please provide as much information as possible about what your PR aims
to do.
PRs with no description will most likely be closed until more
information is provided.
If you're planing on changing fundamental behaviour or add big new
features, please open a GitHub Issue first before starting to work on
it.
If it's not something big and you still want to contact us about it,
feel free to do so !
-->
### Problem description
<!-- Describe the bug that you fixed/feature request that you
implemented, or link to an existing issue describing it -->
This might fix building with capstone 4 as discussed in
https://discord.com/channels/789833418631675954/1155669027306340393/1155669027306340393
### Implementation description
<!-- Explain what you did to correct the problem -->
moves the max definition inside the if statement that checks for
capstone 5
I noticed the bad score on code factor so I reorganized it to make it
more readable and maintainable. In order to break down the big function
into it much smaller parts I encapsulated all the variables that the
functions need to access in two classes, one for the imgui related
statics and the other for non-static variables.
When writing the smaller functions I was noticed that there was room to
simplify the existing algorithms by writing functions that could be
called by parts that previously shared no code. I tested the changes the
same way I tested the original and it seems to work the same way but
maybe a bit faster. Although it may be possible to further optimize the
present code code factor no longer flags the function at all.
As discussed (many times) on Discord, does the same as the new favorite
tag, but instead allows you to add multiple groups.
Initially, this would cause some insane issues with draw/reset
(apparantly) fighting eachother in the pattern drawer. After a lot of
trial and error, I decided to rewrite the flow that is responsible for
calling reset. Now evaluating patterns is the one to decide when the
reset happens, not the core "game"-loop.
To make sure that draw and reset can never happen at the same time, the
mutex originally used for the favorites has been repurposed. Due to the
restructuring, the mutex in the favorite-task is no longer needed, as
that will only ever kick-off after reset is called and if there are
actually patterns, which can never line up to be accessed on different
threads at the same time.
Last but not least, I noticed that hard crashes could result in your
config file getting overridden. I added a check to prevent that.
Last I issue I can see is that if you use an excessive amount of
favorites/groups, a crash can still happen, but it only happens when you
close the program (occasionally, but unpredictable). Before, this would
happen if you ran the evaluation a second time. I boiled the cause of
the crash down to these lines of code in evaluator.cpp >
patternDestroyed:
```cpp
if (pattern->isPatternLocal()) {
if (auto it = this->m_patternLocalStorage.find(pattern->getHeapAddress()); it != this->m_patternLocalStorage.end()) {
auto &[key, data] = *it;
data.referenceCount--;
if (data.referenceCount == 0)
this->m_patternLocalStorage.erase(it);
} else if (!this->m_evaluated) {
err::E0001.throwError(fmt::format("Double free of variable named '{}'.", pattern->getVariableName()));
}
}
```
Specifically, trying to access the `*it` is the reason for the crash
(this was also the cause of the crashes before my fixes, but then during
evaluation).
I'm suspecting the root cause is somewhere in the `.clone` methods of
the patterns. I'd say that for now a crash when closing the program is
more acceptable than during evaluation (which can even happen if you use
favorites).
This PR it just a hack to fix#1160 , it doesn't solve the underlying
problem.
It fixes the problem because by using execvp() directly, it avoids the
call to `sh` done with `system()`, which has a bug on Ubuntu 22.04 which
makes it i,compatibles with the glibc inside the AppImage.
It doesn't fix the underlying problem because the programs we call
themselves still link to the AppImage's libraries instead of the system
ones.
### Problem description
Currently, the providers use the method `isSavable()` to determine both
if they can use "Save" or "Save as".
This behaviour is problematic because some providers may need to be
saveable but not saveable as: for example the view provider. The
original provider may not allow to be saved.
### Implementation description
I separate these two behaviour by creating another function:
`isDumpable()`, that return true by default but can be overridden by the
provider to return false, if the provider should not be dumped in any
way.
### Additional things
While I was at it, I also marked "export" operations as needing the
"dumpable" flag. That way, we can't accidentally export the whole
address space of a process as base64.
I also added documentation for these some functions in Provider
This PR fixes some things about crash handling:
- when the terminate handler is called, immediately set it back to the
original one, so can't make a recursion if the crash-handling code fails
- Only save projects if the crash occured after Imhex finished startup
- do not update the project location when saving the crash backup file:
this will remove problems when `EventAbnormalTermination` is called
before `crashCallback()`
I also added a bit more documentation
