This moves the dumping logic from the GDBRemoteCommunicationHistory
class into the GDBRemotePacket so that it can be reused from the
reproducer command object.
llvm-svn: 372028
To support dumping the reproducer's GDB remote packets, we need the
(de)serialization logic to live in Utility rather than the GDB remote
plugin. This patch renames StreamGDBRemote to GDBRemote and moves the
relevant packet code there.
Its uses in the GDBRemoteCommunicationHistory and the
GDBRemoteCommunicationReplayServer are updated as well.
Differential revision: https://reviews.llvm.org/D67523
llvm-svn: 371907
The StringRef should always be identical to the C string, so we
might as well just create the StringRef from the C-string. This
might be slightly slower until we implement the storage of ArgEntry
with a string instead of a std::unique_ptr<char[]>. Until then we
have to do the additional strlen on the C string to construct the
StringRef.
llvm-svn: 371842
This code is not on any performance critical path that would
justify this shortening optimization. It also makes it possible
to turn 'ref' into a function (as this is the only place where
we modify this ArgEntry member).
llvm-svn: 371836
This just moves the CommandLoader utility into the reproducer namespace
and makes it accessible outside the API layer. This is setting things up
for a bigger change.
llvm-svn: 371689
Originally the idea was for providers to be defined close to where they
are used. While this helped designing the providers in such a way that
they don't depend on each other, it also means that it's not possible to
access them from a central place. This proved to be a problem for some
providers and resulted in them living in the reproducer class.
The ProcessGDBRemote provider is the last remaining exception. This
patch makes things consistent and moves it into the reproducer like the
other providers.
llvm-svn: 371685
This patch adds basic support for DW_OP_convert[1] for integer
types. Recent versions of LLVM's optimizer may insert this opcode into
DWARF expressions. DW_OP_convert is effectively a type cast operation
that takes a reference to a base type DIE (or zero) and then casts the
value at the top of the DWARF stack to that type. Internally this
works by changing the bit size of the APInt that is used as backing
storage for LLDB's DWARF stack.
I managed to write a unit test for this by implementing a mock YAML
object file / module that takes debug info sections in yaml2obj
format.
[1] Typed DWARF stack. http://www.dwarfstd.org/ShowIssue.php?issue=140425.1
<rdar://problem/48167864>
Differential Revision: https://reviews.llvm.org/D67369
llvm-svn: 371532
Summary:
We still have some leftovers of the old completion API in the internals of
LLDB that haven't been replaced by the new CompletionRequest. These leftovers
are:
* The return values (int/size_t) in all completion functions.
* Our result array that starts indexing at 1.
* `WordComplete` mode.
I didn't replace them back then because it's tricky to figure out what exactly they
are used for and the completion code is relatively untested. I finally got around
to writing more tests for the API and understanding the semantics, so I think it's
a good time to get rid of them.
A few words why those things should be removed/replaced:
* The return values are really cryptic, partly redundant and rarely documented.
They are also completely ignored by Xcode, so whatever information they contain will end up
breaking Xcode's completion mechanism. They are also partly impossible to even implement
as we assign negative values special meaning and our completion API sometimes returns size_t.
Completion functions are supposed to return -2 to rewrite the current line. We seem to use this
in some untested code path to expand the history repeat character to the full command, but
I haven't figured out why that doesn't work at the moment.
Completion functions return -1 to 'insert the completion character', but that isn't implemented
(even though we seem to activate this feature in LLDB sometimes).
All positive values have to match the number of results. This is obviously just redundant information
as the user can just look at the result list to get that information (which is what Xcode does).
* The result array that starts indexing at 1 is obviously unexpected. The first element of the array is
reserved for the common prefix of all completions (e.g. "foobar" and "footar" -> "foo"). The idea is
that we calculate this to make the life of the API caller easier, but obviously forcing people to have
1-based indices is not helpful (or even worse, forces them to manually copy the results to make it
0-based like Xcode has to do).
* The `WordComplete` mode indicates that LLDB should enter a space behind the completion. The
idea is that we let the top-level API know that we just provided a full completion. Interestingly we
`WordComplete` is just a single bool that somehow represents all N completions. And we always
provide full completions in LLDB, so in theory it should always be true.
The only use it currently serves is providing redundant information about whether we have a single
definitive completion or not (which we already know from the number of results we get).
This patch essentially removes `WordComplete` mode and makes the result array indexed from 0.
It also removes all return values from all internal completion functions. The only non-redundant information
they contain is about rewriting the current line (which is broken), so that functionality was moved
to the CompletionRequest API. So you can now do `addCompletion("blub", "description", CompletionMode::RewriteLine)`
to do the same.
For the SB API we emulate the old behaviour by making the array indexed from 1 again with the common
prefix at index 0. I didn't keep the special negative return codes as we either never sent them before (e.g. -2) or we
didn't even implement them in the Editline handler (e.g. -1).
I tried to keep this patch minimal and I'm aware we can probably now even further simplify a bunch of related code,
but I would prefer doing this in follow-up NFC commits
Reviewers: JDevlieghere
Reviewed By: JDevlieghere
Subscribers: arphaman, abidh, lldb-commits
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D66536
llvm-svn: 369624
The implementation of this function was obviously incorrect, as the
result variable was never used. This led me to check if it was actually
used anywhere, which came back negative.
llvm-svn: 369492
While generating the Doxygen I noticed this lone namespace that has one
class and one function in it. This moves them into lldb_private.
llvm-svn: 369485
I find as a good cleanup to drop the Compile method. As I do not find TIMTOWTDI
as an advantage and there is already constructor parameter to compile the
regex.
Differential Revision: https://reviews.llvm.org/D66392
llvm-svn: 369352
Originally I wanted to remove the RegularExpression class in Utility and
replace it with llvm::Regex. However, during that transition I noticed
that there are several places where need the regular expression string.
So instead I propose to keep the RegularExpression class and make it a
thin wrapper around llvm::Regex.
This patch also removes the workaround for empty regular expressions.
The result is that we are now (more or less) POSIX conformant.
Differential revision: https://reviews.llvm.org/D66174
llvm-svn: 369153
Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.
Differential revision: https://reviews.llvm.org/D66259
llvm-svn: 368933
Completion requests have two fields that are essentially unimplemented:
`m_match_start_point` and `m_max_return_elements`. This would've been
okay, if it wasn't for the fact that this caused a bunch of useless
parameters to be passed around. Occasionally there would be a comment or
assert saying that they are not supported. This patch removes them.
llvm-svn: 367385
When investigating a completion bug I got confused by the API.
LongestCommonPrefix finds the longest common prefix of the strings in
the string list. Instead of returning that string through an output
argument, just return it by value.
llvm-svn: 367384
The file collector class is useful for creating reproducers,
not just for LLDB, but for other tools as well in LLVM/Clang.
Differential Revision: https://reviews.llvm.org/D65237
llvm-svn: 366956
This patch replaces explicit calls to log::Printf with the new LLDB_LOGF
macro. The macro is similar to LLDB_LOG but supports printf-style format
strings, instead of formatv-style format strings.
So instead of writing:
if (log)
log->Printf("%s\n", str);
You'd write:
LLDB_LOG(log, "%s\n", str);
This change was done mechanically with the command below. I replaced the
spurious if-checks with vim, since I know how to do multi-line
replacements with it.
find . -type f -name '*.cpp' -exec \
sed -i '' -E 's/log->Printf\(/LLDB_LOGF\(log, /g' "{}" +
Differential revision: https://reviews.llvm.org/D65128
llvm-svn: 366936
This patch removes any remaining instances of LogIfAnyCategoriesSet and
replaces them with the LLDB_LOG macro. This in turn made it possible to
make Log::VAPrintf and Log::VAError private.
llvm-svn: 366768
We had a long discussion in D59911 about lldb_assert and what it means.
The result was the assert manifesto on lldb.llvm.org.
> LLDB provides lldb_assert() as a soft alternative to cover the middle
> ground of situations that indicate a recoverable bug in LLDB. In a
> Debug configuration lldb_assert() behaves like assert(). In a Release
> configuration it will print a warning and encourage the user to file a
> bug report, similar to LLVM’s crash handler, and then return
> execution.
However, currently lldb_assert doesn't behave they way it's being
described there: it doesn't abort in a debug/assert build. This patch
fixes that by adding a call to assert() in lldb_assert().
Differential revision: https://reviews.llvm.org/D64267#1571962
llvm-svn: 365246
Copy over access and modification time for the files included in the
reproducer. This is needed to pass tests that check the integrity of
object files based on their time stamp.
llvm-svn: 364457
Before this patch, reproducers weren't relocatable. The reproducer
contained hard coded paths in the VFS mapping, as well in the yaml file
listing the different input files for the command interpreter. This
patch changes that:
- Use relative paths for the DataCollector.
- Use an overlay prefix for the FileCollector.
Differential revision: https://reviews.llvm.org/D63467
llvm-svn: 363697
Utility doesn't link against lldbBase so we cannot call GetVersion in
keep. I already added a string member m_version to deal with that, but
the call was still there.
llvm-svn: 363228
Generally, reproducers are rev-locked to the version of LLDB, so it's
valuable to have the LLDB version in the reproducer. For now I just want
the information to be present, without enforcing it, but I envision
emitting a warning during replay in the future.
Differential revision: https://reviews.llvm.org/D63229
llvm-svn: 363225
This replaces the `info` typedef with a nested struct named Info. This
means we now have FooProvider and FooProvider::Info, instead of two
related but separate classes FooProvider and FooInfo. This change is
mostly cosmetic.
llvm-svn: 363211
Summary:
This is the first of a few patches I have to improve the performance of dynamic module loading on Android.
In this first diff I'll describe the context of my main motivation and will then link to it in the other diffs to avoid repeating myself.
## Motivation
I have a few scenarios where opening a specific feature on an Android app takes around 40s when lldb is attached to it. The reason for that is because 40 modules are dynamicly loaded at that point in time and each one of them is taking ~1s.
## The problem
To learn about new modules we have a breakpoint on a linker function that is called twice whenever a module is loaded. One time just before it's loaded (so lldb can check which modules are loaded) and another right after it's loaded (so lldb can check again which ones are loaded and calculate the diference).
It's figuring out which modules are loaded that is taking quite some time. This is currently done by traversing the linked list of loaded shared libraries that the linker maintains in memory. Each item in the linked list requires its own `x` packet sent to the gdb server (this is android so the network also plays a part). In my scenario there are 400+ loaded libraries and even though we read 0x800 worth of bytes at a time we still make ~180 requests that end up taking 150-200ms.
We also do this twice, once before the module is loaded (state = eAdd) and another right after (state = eConsistent) which easly adds up to ~400ms per module.
## A solution
**Implement `xfer:libraries-svr4` in lldb-server:**
I noticed in the code that loads the new modules that it had support for the `xfer:libraries-svr4` packet (added ~4 years ago to support the ds2 debug server) but we didn't support it in lldb-server. This single packet returns an xml list of all the loaded modules by the process. The advantage is that there's no more need to make 180 requests to read the linked list. Additionally this new requests takes around 10ms.
**More efficient usage of the `xfer:libraries-svr4` packet in lldb:**
When `xfer:libraries-svr4` is available the Process class has a `LoadModules` function that requests this packet and then loads or unloads modules based on the current list of loaded modules by the process.
This is the function that is used by the DYLDRendezvous class to get the list of loaded modules before and after the module is loaded. However, this is really not needed since the LoadModules function already loaded or unloaded the modules accordingly. I changed this strategy to call LoadModules only once (after the process has loaded the module).
**Bugs**
I found a few issues in lldb while implementing this and have submitted independent patches for them.
I tried to devide this into multiple logical patches to make it easier to review and discuss.
## Tests
I wanted to put these set of diffs up before having all the tests up and running to start having them reviewed from a techical point of view. I'm also having some trouble making the tests running on linux so I need more time to make that happen.
# This diff
The `xfer` packages follow the same protocol, they are requested with `xfer:<object>:<read|write>:<annex>:<offset,length>` and a return that starts with `l` or `m` depending if the offset and length covers the entire data or not. Before implementing the `xfer:libraries-svr4` I refactored the `xfer:auxv` to generically handle xfer packets so we can easly add new ones.
The overall structure of the function ends up being:
* Parse the packet into its components: object, offset etc.
* Depending on the object do its own logic to generate the data.
* Return the data based on its size, the requested offset and length.
Reviewers: clayborg, xiaobai, labath
Reviewed By: labath
Subscribers: mgorny, krytarowski, lldb-commits
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D62499
llvm-svn: 362982
Problem discovered in the breakpoint lit test, but probably exists in others.
lldb-test splits lines on LF. Input files that are CR+LF separated (as is
common on Windows) then resulted in commands being sent to LLDB that ended
in CR, which confused the command interpreter.
This could be fixed at different levels:
1. Treat '\r' like a tab or space in the argument splitter.
2. Fix the line splitters (plural) in lldb-test.
3. Normalize the test files to LF only.
If we did only 3, I'd expect similar problems to recur, so this patch does
1 and 2. I may also do 3 in a separate patch later, but that's tricky
because I believe we have some input files that MUST use CR+LF.
Differential Revision: https://reviews.llvm.org/D62759
llvm-svn: 362844
Summary:
The `log timer dump` is showing the time of the function itself minus any function that is called from this one that also happens to be timed. However, this is really not obvious and it also makes it hard to understand the time spent in total and also which children are actually taking the time.
To get a better reading of the timer dump I added the total, children (which I named child) and also the hit count. I used these timers to figure out a performance issue and only after adding this things were more clear to me.
It looks like this:
```
(lldb) log timer dump
35.447713617 sec (total: 35.449s; child: 0.001s; count: 1374) for void SymbolFileDWARF::Index()
29.717921481 sec (total: 29.718s; child: 0.000s; count: 8230500) for const lldb_private::ConstString &lldb_private::Mangled::GetDemangledName(lldb::LanguageType) const
21.049508865 sec (total: 24.683s; child: 3.633s; count: 1399) for void lldb_private::Symtab::InitNameIndexes()
...
```
Reviewers: clayborg, teemperor, labath, espindola, xiaobai
Reviewed By: labath, xiaobai
Subscribers: emaste, mgorny, arichardson, eraman, MaskRay, jdoerfert, labath, davide, teemperor, aprantl, erik.pilkington, jfb, abidh, lldb-commits
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D61235
llvm-svn: 361987
I touched the Broadcaster class earlier today (r361544) and noticed a
few things that could be improved. This patch includes variety of small
fixes: use early returns, use LLDB_LOG macro, use doxygen comments and
finally format the class.
llvm-svn: 361597
The copy constructor of RegularExpression doesn't initialize m_comp_err. This causes an use-of-initialized-value error when a RegularExpression is copied: the copy constructor calls Compile, which calls Free to free the existing regex if needed, which in turn reads m_comp_err to check if there's any regex to be freed.
This change calls the default constructor from the other constructors to make sure members are always initialized with sensible values. This also avoids duplicating init logic, like the RegularExpression(llvm:StringRef) constructor does, which is error prone.
Differential Revision: https://reviews.llvm.org/D62334
llvm-svn: 361546
Jonas Devlieghere