Until there's a "Wait", there's usually only one API message inflight at
a time.
In our quest for performance, we put that single API message in charge
of its own buffer management: instead of allocating buffers on the heap
and deleting them later (storing pointers to them at the far corners of
the earth), it would instead allocate them from small internal pools (if
possible) and only heap allocate (transparently) if necessary. The
pointers flung to the corners of the earth would be pointers (1) back
into the API_MSG or (2) to a heap block owned by boost::small_vector.
It took us months to realize that those bare pointers were being held by
COOKED_READ and RAW_READ and not actually being updated when the API
message was _copied_ as it was shuffled off to the background to become
a "Wait" message.
It turns out that it's trivially possible to crash the console by
sending two API calls--one that waits and one that completes
immediately--when the waiting message or the "wait completer" has a
bunch of dangling pointers in it. It further turns out that some
accessibility software (like JAWS) attaches directly to the console
session, much like winpty and ConEmu and friends. They're trying to read
out the buffer (API call!) and sometimes there's a shell waiting for
input (API call!). Oops.
In this commit, we fix up the message's internal pointers (in lieu of
giving it a proper copy constructor; see GH-10076) and then tell the
wait completion routine (which is going to be a COOKED_READ, RAW_READ,
DirectRead or WriteData) about the new buffer location.
This is a scoped fix that should be replaced (TODO GH-10076) with a
final one after Ask mode.
Retrieved from https://microsoft.visualstudio.com os.2020 OS official/rs_wdx_dxp_windev eca0875950fd3a9735662474613405e2dc06f485
References GH-10076
Fixes MSFT-33127449
Fixes GH-9692
There is going to be a very long tail of applications that will
explicitly request VT SGR 40/37 when what they really want is to
SetConsoleTextAttribute() with a black background/white foreground.
Instead of making those applications look bad (and therefore making us
look bad, because we're releasing this as an update to something that
"looks good" already), we're introducing this compatibility quirk.
Before the color reckoning in #6698 + #6506, *every* color was subject
to being spontaneously and erroneously turned into the default color.
Now, only the 16-color palette value that matches the active console
background/foreground color will be destroyed, and only when received
from specific applications.
Removal will be tracked by #6807.
Michael and I discussed what layer this quirk really belonged in. I
originally believed it would be sufficient to detect a background color
that matched the legacy default background, but @j4james provided an
example of where that wouldn't work out (powershell setting the
foreground color to white/gray). In addition, it was too heavyhanded: it
re-broke black backgrounds for every application.
Michael thought that it should live in the server, as a small VT parser
that righted the wrongs coming directly out of the application. On
further investigation, however, I realized that we'd need to push more
information up into the server (so that it could make the decision about
which VT was wrong and which was right) than should be strictly
necessary.
The host knows which colors are right and wrong, and it gets final say
in what ends up in the buffer.
Because of that, I chose to push the quirk state down through
WriteConsole to DoWriteConsole and toggle state on the
SCREEN_INFORMATION that indicates whether the colors coming out of the
application are to be distrusted. This quirk _only applies to pwsh.exe
and powershell.exe._
NOTE: This doesn't work for PowerShell the .NET Global tool, because it
is run as an assembly through dotnet.exe. I have no opinion on how to
fix this, or whether it is worth fixing.
VALIDATION
----------
I configured my terminals to have an incredibly garish color scheme to
show exactly what's going to happen as a result of this. The _default
terminal background_ is purple or red, and the foreground green. I've
printed out a heap of test colors to see how black interacts with them.
Pull request #6810 contains the images generated from this test.
The only color lines that change are the ones where black as a
background or white as a foreground is selected out of the 16-color
palette explicitly. Reverse video still works fine (because black is in
the foreground!), and it's even possible to represent "black on default"
and reverse it into "default on black", despite the black in question
having been `40`.
Fixes#6767.
