| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Useful when installed as shared libraries, so other viewer executables
can share these libraries.
|
|
by removing cyclic dependencies, and allowing shlib undefined on
SLPlugin alone.
|
|
|
|
when compiling newview using GCC.
|
|
Streaming is not working yet, though. Until it's made sure that the
dynamic library and plugins needed are on the paths the executable is
expecting them to be.
|
|
This reverts commit f4c8949ac66d08263845f60a7cef2ecb9c77079b.
|
|
System libcurl, which is typically newer, doesn't accept when SL server
responses with an invalid Content-Encoding value (usually some value
that's probably meant to be put as the Content-Type value), that we'd
get "unrecognized or bad HTTP Content or Transfer-Encoding"
https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Accept-Encoding
A way to fix this would be to just not expect decompressed contents, by
letting libcurl have the default value for CURLOPT_ACCEPT_ENCODING, which
is NULL.
|
|
Even though the account was logged in, it would get stuck at getting
region handshake. The problem was because the viewer wasn't getting the
acknowledgement to the successfully sent UseCircuitCode message. I
compared the message data, and it differed (from Linux) on the byte
order of the Code variable (the SessionID & agent ID were right). The
bytes sent to the network weren't reversed (and I was on an Intel
processor).
|
|
Though without this, the viewer had still successfully built, and I
didn't experience any run-time problem yet. This commit is to anticipate
any directory picker related problem later, cause it seems very likely
that this is needed.
|
|
|
|
instead of letting it fallback to the default which would be Window's.
When using the default, somehow the viewer launched with no colours even
after resetting ~/.secondlife/user_settings/settings.xml.
|
|
|
|
The function takes a boolean argument anyway. This is so we don't get
GCC int in bool context warning which would be treated as an error.
|
|
Otherwise GCC would treat them as errors, if not suppressed.
|
|
|
|
|
|
in order to get rid of undefined references to
`LLFilePicker::getOpenFileModeless(LLFilePicker::ELoadFilter, void (*)(bool, std::vector<std::string, std::allocator<std::string> >&, void*), void*)'
`LLFilePicker::getMultipleOpenFilesModeless(LLFilePicker::ELoadFilter, void (*)(bool, std::vector<std::string, std::allocator<std::string> >&, void*), void*)'
`LLFilePicker::getSaveFileModeless(LLFilePicker::ESaveFilter, std::string const&, void (*)(bool, std::string&, void*), void*)'
The UI has been relying on modeless file operations. UI implementations
for Linux would fall within the GTK scope, and there haven't been
implementations for these three methods yet. Even know they're
defined using member functions that do nothing, and return boolean
false.
|
|
|
|
Mostly following Linux.
|
|
On some platforms,
pkg-config --libs vorbis
might not necessarily imply -logg.
vorbisenc & vorbisfile need their own checks anyway.
|
|
Including v4 would cause conflicts.
|
|
|
|
For now. Maybe.
|
|
Otherwise on GCC the -Waddress & -Wno-nonnull-compare warnings would be
treated as errors. It's a different term on Clang, which is undefined
bool conversion. On Clang it's just a warning, but on AppleClang it's
treated as an error.
Rather than suppressing these warnings, I think it's better to just
adjust the interface of copyTexture, since it isn't called that many
times.
|
|
Otherwise we'd get this.
error: declaration of anonymous struct must be a definition
struct Status
^
indra/llcommon/llprocess.h:284:2: warning:
declaration does not declare anything [-Wmissing-declarations]
|
|
|
|
|
|
|
|
When using system vanilla COLLADA DOM, it still uses std::auto_ptr which
is deprecated in C++17.
|
|
which can be found too in the system COLLADA DOM headers.
|
|
cause they would cause ambiguity with system COLLADA DOM classes.
|
|
as it's not implied on some platforms that std::array would be
unrecognised.
|
|
The style conventions aren't really being followed that the different
styles of using tabs or spaces as indentations lead to GCC considering
them as misleading.
It's better to just fix them (but as little as possible as to minimise
this fork difference from upstream) than to supress the warnings from
being treated as errors.
|
|
On GCC, compiling against a recent GTK2 version would stop on deprecated
pre-processors.
|
|
and any Linux SDL code should be applied to FreeBSD too.
|
|
Otherwise _1 and _2 would be considered undeclared in many places.
|
|
|
|
in order to get rid of this:
error: invalid new-expression of abstract class type ‘LLWindowSDL’
fullscreen, clearBg, enable_vsync, use_gl, ignore_pixel_depth, fsaa_samples);
note: because the following virtual functions are pure within ‘LLWindowSDL’:
class LLWindowSDL : public LLWindow
^~~~~~~~~~~
virtual void* createSharedContext() = 0;
^~~~~~~~~~~~~~~~~~~
virtual void makeContextCurrent(void* context) = 0;
^~~~~~~~~~~~~~~~~~
virtual void destroySharedContext(void* context) = 0;
^~~~~~~~~~~~~~~~~~~~
virtual void toggleVSync(bool enable_vsync) = 0;
^~~~~~~~~~~
The window has been relying on some, mostly GL context related, methods.
These methods are declared abstract, so for now they're implemented
using member functions that do nothing, return nothing.
|
|
X11 is still required too, that's why it's still added for linking.
On Darwin, SDL is disabled for now.
|
|
and only on a specific file.
|
|
For now the directory name is hardcoded since v2.5.0 is the latest and the viewer code already depends on a certain function from that version. So for now it's safe to assume that the directory name's suffix is 2.5.
|
|
|
|
From the errors:
base type ‘float (*)(float, float) noexcept’ fails to be a struct or class type
struct functor_action : public FuncT {
^~~~~~~~~~~~~~
‘float (*)(float, float) noexcept’ is not a class, struct, or union type
struct result { typedef typename FuncT::result_type type; };
^~~~
type ‘float (*)(float, float) noexcept’ is not a direct base of
‘phoenix::functor_action<float (*)(float, float) noexcept>’
: FuncT(fptr_) {}
^
it seems that GCC, at least the ones on my system, failed to deduce that
the binder meant to be used when trying to bind the C Std Math Library's
powf and fmodf functions is of course the function pointer one
(specifically the 2 args specialisation), and not the functor one.
|
|
since Meshoptimizer v0.18, an option to specify meshopt_SimplifyLockBorder.
|
|
|
|
|
|
|
|
Relies on sysctl, like on Darwin, and on parsing a file, like on Linux,
except the file would be /var/run/dmesg.boot.
|
|
Since C++11, NULL is promoted to nullptr on some BSD platforms. This is
very problematic when used with Boost. At times it would fail during
compile-time. What's worse is if it passes compile-time, but then crash
during run-time, for example when some condition is to be checked for its
truth, when then it would be compared to a nullptr.
|
