Summary:
The DWARF spec states that the DWARF stack arguments are numbered from
the top. Our implementation of DW_OP_pick was counting them from the
bottom.
This bug probably wasn't noticed because nobody (except my upcoming
postfix-to-DWARF converter) uses DW_OP_pick, but I've cross-checked with
gdb to confirm that counting from the top is the expected behavior.
This patch fixes the implementation to match the spec and gdb behavior
and adds a test.
Reviewers: jasonmolenda, clayborg
Subscribers: mgorny, aprantl, lldb-commits
Differential Revision: https://reviews.llvm.org/D61182
llvm-svn: 359436
We recently moved API logging into the instrumentation macros. This made
that logging is now consistent and abstracted behind a macro for every
API functions, independent of the reproducers. It also means we have a
lot more output. While this is a good thing, it also meant a lot more
noise in the log, from things that aren't always equally interesting,
such as the copy constructor for example.
To improve usability, we should increase the signal-to-noise ratio. I
propose to achieve this by only logging API functions that cross the API
boundary. This is a divergence of what we had before, where a select
number of functions were logged, irregardless of the API boundary, a
concept that was introduced for the reproducers. However, I believe this
is in line with the purpose of the API log.
Differential revision: https://reviews.llvm.org/D60984
llvm-svn: 359016
Deallocating the data recorder in during the ::Keep() operation causes
problems down the line when exiting the debugger. The command
interpreter still holds a pointer to the now deallocated object and has
no way to know it no longer exists. This is exactly what the m_record
flag was meant for, although it wasn't hooked up properly either.
llvm-svn: 358916
There is an alternative method to GetConstCStringWithLength that
takes a StringRef. GetConstCStringWithLength also calls this
method in the end, so directly calling the StringRef saves
us from a unnecessary conversion to a C-string.
llvm-svn: 358357
A lot of comments in LLDB are surrounded by an ASCII line to delimit the
begging and end of the comment.
Its use is not really consistent across the code base, sometimes the
lines are longer, sometimes they are shorter and sometimes they are
omitted. Furthermore, it looks kind of weird with the 80 column limit,
where the comment actually extends past the line, but not by much.
Furthermore, when /// is used for Doxygen comments, it looks
particularly odd. And when // is used, it incorrectly gives the
impression that it's actually a Doxygen comment.
I assume these lines were added to improve distinguishing between
comments and code. However, given that todays editors and IDEs do a
great job at highlighting comments, I think it's worth to drop this for
the sake of consistency. The alternative is fixing all the
inconsistencies, which would create a lot more churn.
Differential revision: https://reviews.llvm.org/D60508
llvm-svn: 358135
For some reason I had convinced myself that functions returning by
pointer or reference do not require recording their result. However,
after further considering I don't see how that could work, at least not
with the current implementation. Interestingly enough, the reproducer
instrumentation already (mostly) accounts for this, though the
lldb-instr tool did not.
This patch adds the missing macros and updates the lldb-instr tool.
Differential revision: https://reviews.llvm.org/D60178
llvm-svn: 357639
The utility library shouldn't depend on curses, libedit or python. Move
curses to core, libedit to host and python to the python plugin.
Differential revision: https://reviews.llvm.org/D59970
llvm-svn: 357287
FindPythonInterp and FindPythonLibs do two things, they set some
variables (PYTHON_LIBRARIES, PYTHON_INCLUDE_DIRS) and update the cached
variables (PYTHON_LIBRARY, PYTHON_INCLUDE_DIR) which are also used to
specify a custom python installation.
I believe the canonical way to do this is to use the PYTHON_LIBRARIES
and PYTHON_INCLUDE_DIRS variables instead of the cached ones. However,
since the cached variables are accessible from the cache and GUI, this
is a lot less confusing when you're trying to debug why a variable did
or didn't get the value you expected. Furthermore, as far as I can tell,
the implementation uses the cached variables to set their LIBRARIES/DIRS
counterparts. This is also the reason this works today even though we
mix-and-match.
Differential revision: https://reviews.llvm.org/D59968
llvm-svn: 357282
Currently LLDB crashes when autocompleting a command that ends with a
backtick because the quote character wasn't handled. This fixes that and
adds a unit test for this function.
Differential revision: https://reviews.llvm.org/D59779
llvm-svn: 356927
The command interpreter holds a pointer to a DataRecorder. After
generating the reproducer, we deallocated all the DataRecorders, causing
the command interpreter to hold a non-null reference to an invalid
object.
This patch changes the behavior of the command provider to stop the
DataRecorders when a reproducer is generated, rather than deallocating
them.
llvm-svn: 355940
My apologies for the large patch. With the exception of ConstString.h
itself it was entirely produced by sed.
ConstString has exactly one const char * data member, so passing a
ConstString by reference is not any more efficient than copying it by
value. In both cases a single pointer is passed. But passing it by
value makes it harder to accidentally return the address of a local
object.
(This fixes rdar://problem/48640859 for the Apple folks)
Differential Revision: https://reviews.llvm.org/D59030
llvm-svn: 355553
There are set of classes in Target that describe the parameters of a
process - e.g. it's PID, name, user id, and similar. However, since it
is a bare description of a process and contains no actual functionality,
there's nothing specifically that makes this appropriate for being in
Target -- it could just as well be describing a process on the host, or
some hypothetical virtual process that doesn't even exist.
To cement this, I'm moving these classes to Utility. It's possible that
we can find a better place for it in the future, but as it is neither
Host specific nor Target specific, Utility seems like the most appropriate
place for the time being.
After this there is only 2 remaining references to Target from Host,
which I'll address in a followup.
Differential Revision: https://reviews.llvm.org/D58842
llvm-svn: 355342
Summary:
This creates an abstract base class called "UserIDResolver", which can
be implemented to provide user/group ID resolution capabilities for
various objects. Posix host implement a PosixUserIDResolver, which does
that using posix apis (getpwuid and friends). PlatformGDBRemote
forwards queries over the gdb-remote link, etc. ProcessInstanceInfo
class is refactored to make use of this interface instead of taking a
platform pointer as an argument. The base resolver class already
implements caching and thread-safety, so implementations don't have to
worry about that.
The main motivating factor for this was to remove external dependencies
from the ProcessInstanceInfo class (so it can be put next to
ProcessLaunchInfo and friends), but it has other benefits too:
- ability to test the user name caching code
- ability to test ProcessInstanceInfo dumping code
- consistent interface for user/group resolution between Platform and
Host classes.
Reviewers: zturner, clayborg, jingham
Subscribers: mgorny, lldb-commits
Differential Revision: https://reviews.llvm.org/D58167
llvm-svn: 355323
automatic move should not fire when returning type T in a function with
result type Expected<T>. Some compilers seem to allow that nonetheless.
llvm-svn: 355270
This patch adds the necessary logic to capture and replay commands
entered into the command interpreter. A DataRecorder shadows the input
and writes its data to a know file. During replay this file is used as
the command interpreter's input.
It's possible to the command interpreter more than once, with a
different input source. We support this scenario by using multiple
buffers. The synchronization for this takes place at the SB layer, where
we create a new recorder every time the debugger input is changed.
During replay we use the corresponding buffer as input.
Differential revision: https://reviews.llvm.org/D58564
llvm-svn: 355249
Debugging issues with instrumentation capture and replay can be
particularly tricky, especially because part of the process takes places
even before the debugger is initialized. This patch adds more logging
capabilities to these classes, hidden behind a macro define.
Differential revision: https://reviews.llvm.org/D58566
llvm-svn: 355002
Summary:
This behavior was originally added in rL252264 (git commit 76a7f365da)
in order to be extra careful with handling platforms like watchos and tvos.
However, as far as triples go, those two (and others) are treated as OSes and
not environments, so that should not really apply here.
Additionally, this behavior is incorrect and can lead to incorrect ArchSpecs.
Because android is specified as an environment and not an OS, not propogating
the environment can lead to modules and targets being misidentified.
Differential Revision: https://reviews.llvm.org/D58664
llvm-svn: 354938
Summary:
These functions should always return the opposite of the
`Triple{Environment,OS,Vendor}WasSpecified` functions. Unspecified unknown is
the same as unspecified, which is why one set of functions should give us what
we want. It's possible to have specified unknown, which is why we can't just
rely on checking the enum values of vendor/os/environment. We must also ensure
that the names of these are empty and not "unknown".
Differential Revision: https://reviews.llvm.org/D58653
llvm-svn: 354933
As per the discussion on the mailing list:
http://lists.llvm.org/pipermail/lldb-commits/Week-of-Mon-20190218/048007.html
This commit implements option (3):
> Go back to initializing the reproducer before the rest of the debugger.
> The method wouldn't be instrumented and guarantee no other SB methods are
> called or SB objects are constructed. The initialization then becomes part
> of the replay.
Differential revision: https://reviews.llvm.org/D58410
llvm-svn: 354631
This enables the function to be called with a StringRef without jumping
through any hoops. I rename the function to "PutStringAsRawHex8" to
honor the extended interface. I also remove ".c_str()" from any calls to
this function I could find.
llvm-svn: 353841
Summary:
This adds support for auto-detection of path style to SymbolFileBreakpad
(similar to how r351328 did the same for DWARF). We guess each file
entry separately, as we have no idea which file came from which compile
units (and different compile units can have different path styles). The
breakpad generates should have already converted the paths to absolute
ones, so this guess should be reasonable accurate, but as always with
these kinds of things, it is hard to give guarantees about anything.
In an attempt to bring some unity to the path guessing logic, I move the
guessing logic from inside SymbolFileDWARF into the FileSpec class and
have both symbol files use it to implent their desired behavior.
Reviewers: clayborg, lemo, JDevlieghere
Subscribers: aprantl, markmentovai, lldb-commits
Differential Revision: https://reviews.llvm.org/D57895
llvm-svn: 353702
GetIndexForObjectImpl generated a bunch of "conversion casts away
constness warnings". Change the function to use "const void *" (and
static_cast, while I'm at it), to avoid this.
Driver.cpp: unused variable "replay" (this was actually caused by a
subsequent partial revert of this patch). I just finish the revert by
removing the variable completely.
llvm-svn: 353405
LLDB testsuite fails when built by GCC8 on:
LLDB :: SymbolFile/DWARF/find-basic-namespace.cpp
This is because this code in LLDB codebase has undefined behavior:
#include <algorithm>
#include <string.h>
// lldb/source/Plugins/ObjectFile/Mach-O/ObjectFileMachO.cpp:1731
static struct section_64 {
char sectname[16];
char segname[16];
} sect64 = { {'_','_','a','p','p','l','e','_','n','a','m','e','s','p','a','c'}, "__DWARF" };
int main() {
return std::min<size_t>(strlen(sect64.sectname), sizeof(sect64.sectname));
}
It has been discussed as a (false) bugreport to GCC:
wrong-code: LLDB testcase fails: SymbolFile/DWARF/find-basic-namespace.cpp
https://bugzilla.redhat.com/show_bug.cgi?id=1672436
Differential Revision: https://reviews.llvm.org/D57781
llvm-svn: 353280
This patch adds the file provider which is responsible for capturing
files used by LLDB.
When capturing a reproducer, we use a file collector that is very
similar to the one used in clang. For every file that we touch, we add
an entry with a mapping from its virtual to its real path. When we
decide to generate a reproducer we copy over the files and their
permission into to reproducer folder.
When replaying a reproducer, we load the VFS mapping and instantiate a
RedirectingFileSystem. The latter will transparently use the files
available in the reproducer.
I've tested this on two macOS machines with an artificial example.
Still, it is very likely that I missed some places where we (still) use
native file system calls. I'm hoping to flesh those out while testing
with more advanced examples. However, I will fix those things in
separate patches.
Differential revision: https://reviews.llvm.org/D54617
llvm-svn: 352538
We use UUID::fromOptionalData to read UUID's from the Mach-O files, so UUID's
of all 0's are invalid UUID's.
We also get uuid's from debugserver, which need to match the file UUID's. So
we need an API that treats "000000000" as invalid as well. Added that and use it.
Differential Revision: https://reviews.llvm.org/D57195
llvm-svn: 352122
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
In the original reproducer design, I expected providers to be more
dynamic than they turned out. For example, we don't have any instances
where one provider has multiple files. Additionally, I expected there to
be less locality between capture and replay, with the provider being
defined in one place and the replay code to live in another. Both
contributed to the design of the provider info.
This patch refactors the reproducer info to be something static. This
means less magic strings and better type checking. The new design still
allows for the capture and replay code to live in different places as
long as they both have access to the new statically defined info class.
I didn't completely get rid of the index, because it is useful for (1)
sanity checking and (2) knowing what files are used by the reproducer.
Differential revision: https://reviews.llvm.org/D56814
llvm-svn: 351501
Summary:
If we opened a file which was produced on system with different path
syntax, we would parse the paths from the debug info incorrectly.
The reason for that is that we would parse the paths as they were
native. For example this meant that on linux we would treat the entire
windows path as a single file name with no directory component, and then
we would concatenate that with the single directory component from the
DW_AT_comp_dir attribute. When parsing posix paths on windows, we would
at least get the directory separators right, but we still would treat
the posix paths as relative, and concatenate them where we shouldn't.
This patch attempts to remedy this by guessing the path syntax used in
each compile unit. (Unfortunately, there is no info in DWARF which would
give the definitive path style used by the produces, so guessing is all
we can do.) Currently, this guessing is based on the DW_AT_comp_dir
attribute of the compile unit, but this can be refined later if needed
(for example, the DW_AT_name of the compile unit may also contain some
useful info). This style is then used when parsing the line table of
that compile unit.
This patch is sufficient to make the line tables come out right, and
enable breakpoint setting by file name work correctly. Setting a
breakpoint by full path still has some kinks (specifically, using a
windows-style full path will not work on linux because the path will be
parsed as a linux path), but this will require larger changes in how
breakpoint setting works.
Reviewers: clayborg, zturner, JDevlieghere
Subscribers: aprantl, lldb-commits
Differential Revision: https://reviews.llvm.org/D56543
llvm-svn: 351328
LLVM added wrappers to std::sort (r327219) that randomly shuffle the
container before sorting. The goal is to uncover non-determinism due to
undefined sorting order of objects having the same key.
This can be enabled with -DLLVM_ENABLE_EXPENSIVE_CHECKS=ON.
llvm-svn: 350679
This patch simplifies boolean expressions acorss LLDB. It was generated
using clang-tidy with the following command:
run-clang-tidy.py -checks='-*,readability-simplify-boolean-expr' -format -fix $PWD
Differential revision: https://reviews.llvm.org/D55584
llvm-svn: 349215
Summary:
These are general purpose "utility" classes, whose functionality is not
debugger-specific in any way. As such, I believe they belong in the
Utility module.
This doesn't break any particular dependency (yet), but it reduces the
number of Core dependencies across the board.
Reviewers: zturner, jingham, teemperor, clayborg
Subscribers: mgorny, lldb-commits
Differential Revision: https://reviews.llvm.org/D55361
llvm-svn: 349157
As Pavel noted on the mailing list we should only create the bottom-most
directory if it doesn't exist. This should also fix the test case on
Windows as we can use lit's temp directory.
llvm-svn: 348289
This patch changes the way the reproducer is initialized. Rather than
making changes at run time we now do everything at initialization time.
To make this happen we had to introduce initializer options and their SB
variant. This allows us to tell the initializer that we're running in
reproducer capture/replay mode.
Because of this change we also had to alter our testing strategy. We
cannot reinitialize LLDB when using the dotest infrastructure. Instead
we use lit and invoke two instances of the driver.
Another consequence is that we can no longer enable capture or replay
through commands. This was bound to go away form the beginning, but I
had something in mind where you could enable/disable specific providers.
However this seems like it adds very little value right now so the
corresponding commands were removed.
Finally this change also means you now have to control this through the
driver, for which I replaced --reproducer with --capture and --replay to
differentiate between the two modes.
Differential revision: https://reviews.llvm.org/D55038
llvm-svn: 348152
When I landed the initial reproducer framework I knew there were some
things that needed improvement. Rather than bundling it with a patch
that adds more functionality I split it off into this patch. I also
think the API is stable enough to add unit testing, which is included in
this patch as well.
Other improvements include:
- Refactor how we initialize the loader and generator.
- Improve naming consistency: capture and replay seems the least ambiguous.
- Index providers by name and make sure there's only one of each.
- Add convenience methods for creating and accessing providers.
Differential revision: https://reviews.llvm.org/D54616
llvm-svn: 347716
Pavel Labath