Turns out that the precautions against winelib builds failing, which I
put in years ago because I was using winelib as a build setup for
Coverity testing, are all obsolete. My Coverity build scripts runs
fine now without any of them.

Aah, that's better. It's been bugging me since I added it.

Many of VS's warnings are too noisy to be useful, but I just tried the
experiment of turning off the unrecoverable ones and seeing what was
left, and I found a couple of things that actually seem worth fixing.

In a few cases in mpint.c, and in one case in sshzlib.c, we had the
idiom 'size_t var = 1 << bitpos;', and VS pointed out that when '1' is
implicitly a 32-bit int and 'size_t' is 64 bits, this is probably not
what you wanted. Writing '(size_t)1 << bitpos' is safer.

Secondly, VS complained about lots of functions failing to return a
value, or not returning a value on every code path. In every case this
was somewhere that we'd used the local unreachable() idiom to indicate
that those code paths didn't return at all. So the real problem was
that that idiom didn't work in VS. And that's not because VS _can't_
mark functions as noreturn: it has a perfectly good declspec for it.
It was just that we hadn't actually _done_ it. Now added a clause in
the #if in defs.h that spots VS and uses the declspec.

In commit d53b3bcd I changed the final setting of kl->broken
so that it wouldn't overwrite a 'true' value set earlier in the
function. But that means it might not be set at all, because I forgot
I now needed to initialise it to false. Ahem.

This replaces the pure radio-button setup that we've always had on the
Session config panel.

Since the last release, that set of radio buttons has been getting out
of hand. We've added two new protocols (SUPDUP, and the 'bare
ssh-connection' aka psusan protocol), neither of which is mainstream
enough to be a sensible thing to wave at all users on the front page
of the config GUI, so that they perhaps start wondering if that's the
protocol they want to use, or get sidetracked by going and looking it
up.

The replacement UI still has radio buttons, but only for the most
common protocols, which will typically be SSH and serial. Everything
else is relegated to a drop-down list sitting next to a third radio
button labelled "Other".

In every be_* module providing a backends[] list, there's also a
variable n_ui_backends which indicates how many of the backends ought
to appear as first-level radio buttons.

(Credit where due: this patch is a joint effort between Jacob and me,
and is one of those rare cases where it would be nice to be able to
put both our names into the Author field of the commit. Failing that,
I can at least mention it here.)

This will let us put two controls side by side (e.g. in disjoint
columns of a multi-col layout) and indicate that instead of the
default behaviour of aligning their top edges, their centreline (or,
even better if available, font baseline) should be aligned.

NFC: nothing uses this yet.

We never expect to be passed a NULL GtkFrontend pointer, and even if
we were, we'd have crashed several lines above this test.

It was benign, of course, but Coverity (which pointed it out) dislikes
this kind of thing on the basis that it's confusing - you ought to
either test it for NULL properly, or not at all - and I see its point.

Coverity points out that it's theoretically possible for the main loop
in radioline_common() to read r.bottom without having gone through the
conditional setup at the start of the function _or_ a previous
iteration of the main loop. I think this can only happen in some silly
case that doesn't actually come up, but on the other hand, it's easy
to add the necessary robustness.

Coverity pointed out that I'd checked if the LoadedFile was NULL, set
an error message ... and then accidentally fallen through to the
success handler anyway.

Coverity points out that if rsa_ssh1_public_blob_len sees data it
doesn't like, it returns -1 to indicate an error. But the code that
uses it to parse the SSH1_AGENT_RSA_IDENTITIES_ANSWER payload was
passing it directly to get_data() as a length field, without checking
for that case. Now we do check it, and use it to set the existing
kl->broken flag that indicates that the key list was not correctly
formatted.

Coverity was unhappy that I'd used the packet length as a loop bound
without sanitising it first (on the basis that it had decided anything
coming from GET_32BIT_MSB_FIRST was potentially tainted).

I think this is not a security issue: all that will happen if the
server sends a huge packet length is that we'll try to allocate space
for it. On a 64-bit machine we might even _succeed_; on 32-bit, we'll
fail, and snewn() will abort the program rather than return NULL. So
*technically* this is a remote-triggered crash. But it can only happen
in a situation where the same server could have triggered the
termination of the SFTP connection just as easily by simply closing it
- the only difference is that the client would die with a different
fatal error message.

(In particular, it isn't even a DoS against other processes
participating in a connection-shared SSH session. The upstream will
pass the SFTP data stream through without parsing it, so it and the
other downstreams will be unaffected. Only the particular downstream
operating the SFTP client will run into this problem.)

If named_pipe_agent_gotdata was called with an error or EOF status, it
would call agent_cancel_query(pq), but then accidentally fall through
to the non-error handler which would dereference pq. I meant to return
early in that situation, and Coverity spotted that I'd left out the
early return statement.

Coverity objected to several similar cases in this code in which I'd
checked a pointer for NULL after already having done things to it. I
think all the cases are benign, in that (as the comments tersely
mention) those checks could only fail if the unifontsel system had got
_really_ confused, in which case probably some other bug would have
been on the point of manifesting anyway. But Coverity has a point
anyway: if I'm _going_ to check those values for NULL, let's check
them consistently.

Commit d851df48 deleted a #if / #else / #endif on the grounds
that the condition would now always be true, without also deleting the
code inside the #else. Happily, the then-branch ended with a return,
so it was a benign mistake - the erroneously left-in else-clause code
was unreachable. But now Coverity has pointed it out, let's remove it.

Coverity points out that we don't need to check the output buffer
bound before writing out the first 32 bytes of each full-length
BLAKE2b invocation, because the only time we're doing a full-length
one in the first place is if the output buffer bound was at least 64
bytes.

(More specifically: whenever we're in the while loop, length > 64, so
setting chunk = 32 and then checking if chunk > length has a totally
predictable answer.)

The winelib headers don't have GWL_foo, only GWLP_foo (which, fair
enough, I should have been using already). And a side effect was to
point out some slightly incautious integer types in printf argument
lists.

This has apparently been missing more or less forever (though Unix
Plink does have it). Without this, ssh.c can't call ldisc_update,
which can't pass the current editing and echoing settings through to
seat_echoedit_update. Windows Plink has always _had_ an implementation
of that seat method (and the static function that preceded it), but it
was never able to be called, because of that missing link.

The result was that manual overrides in the Conf to force local
editing/echoing to a particular state were not honoured by Windows
Plink, and neither were mainchan.c's attempts to set the state
automatically based on whether a pty had been allocated at the far end
of the connection.

We already had one of these in utils, since c18e5dc8.

This seems more useful than the previous behaviour of not prompting for
a passphrase and only emitting the public part; if we want that back
I suppose we could invent a "-O text-public".

Also, document the text dump format a bit in the man page.

Thanks to Jacob for spotting this one: when we hand a passphrase back
to pageant.c via pageant_passphrase_request_success(), if the key
doesn't decrypt successfully, pageant.c responds by immediately
issuing another passphrase prompt - and it does it _synchronously_, by
calling back from within pageant_passphrase_request_success(). In this
case, the effect is that we end up in ask_passphrase_common(), which
starts by asserting that nonmodal_passphrase_hwnd is NULL - but it
wasn't NULL _quite_ yet, because end_passphrase_dialog() was expecting
to clean it up immediately after pageant_passphrase_request_success()
returned, i.e. just too late.

The heavyweight fix would be to arrange a toplevel callback to defer
opening the new window until after the old one had been cleaned up.
But in this case I don't think there's any need: it's enough to simply
do the operations in end_passphrase_dialog() in the opposite order, so
that first we destroy the old window and set nonmodal_passphrase_hwnd
back to NULL, and _then_ we call into pageant.c which might call us
back and open a fresh window.

"-O text" (c18e5dc8) and ability to read key from stdin (a7599a57a3).

(Bringing them in line with the PuTTY one in cc6ab00b.)

Also, ensure -E/--fptype in Unix Pageant is (correctly) documented
everywhere.

Gives more helpful messages if Unix pageant ends up being a client for,
say, OpenSSH's ssh-agent, or indeed an older version of Pageant.

(Also, tweak a couple of other messages that still assumed that
pageant-as-client always talks to Pageant-as-agent.)

The protocol already allowed adding an encrypted form to a cleartext key
already held by the agent, and you might want to do so if, say, the key
happened to originally be added in cleartext-only form but you want to
be able to forget that with 'pageant -R' in future.

Previously this would prevent the About box ever being opened again.

Treat as aborting passphrase input. (Previously it would just hang.)

This was introduced in ca9cd983.

This got broken in 696550a5.

This makes their order match the configuration GUI, as of 18d273fc.

(No change to document content in this commit, just rearrangement.)

Now the Remove button is disabled if there aren't any keys at all
loaded, and the Re-encrypt button is disabled if no key is currently
in a state where it's decrypted but re-encryptable.

Now the systray menu includes 'Remove All Keys' and 'Re-encrypt All
Keys' options, which do exactly what they say on the tin.

Not quite sure how that happened! But at some point in the past, a bunch
of other definitions in winpgnt.c managed to get in between the first
few IDM_FOO constants and the last few. Bring them all back together.