Inhibit suspend during the processing of an imaging operation

[logiclrd]

This pull request is being created as a draft because I have no way to test the code. :-)

• I don't have a Linux or Mac build environment. (I do have a Linux system but it doesn't have a working Qt installation.)
• I tried to configure a Windows build environment but messed up somewhere along the way. It's trying to use CL to build the C code, but it's using GNU extensions so that's not working.

I have executed the Windows and Linux specific inhibitor code in isolated test projects and it builds and seems to work. But, I don't have any way to tell if the other code even compiles. I'm hoping there will be build pipelines that'll take care of at least part of that.

Someone with a Mac will need to test the Mac code.

This PR is a direct response to a poor user experience I had with the tool. Per issue #1188, I set up an operation to download & write an image to a Micro SD card, and then I left the system doing its thing. When I returned, I found out that the system had deemed itself idle and gone to sleep after 15 minutes, long before the download completed.

This PR aims to address this by leveraging OS-specific functionality to inhibit autosuspend.

• On Windows, it sets the thread execution state to include the flag ES_SYSTEM_REQUIRED.
• On OS X, it creates an IOPM assertion of type PreventUserIdleSystemSleep.
• On Linux, it tries three strategies:
  • It attempts to connect to the DBus endpoint org.gnome.SessionManager. If this succeeds, it makes a call to the Inhibit method.
  • It attempts to launch system utility systemd-inhibit.
  • It attempts to launch system utility kde-inhibit.

These -inhibit utilities in the Linux-specific side work by taking a command and executing it in a scope that inhibits whatever action is to be inhibited. The inhibition remains in effect until the process exits. In order to invert this control mechanism, the code here creates a FIFO and then has the command in the -inhibit scope read from it. It can then terminate the read at any time by closing its end of the pipe.

Any errors are silently ignored; a best effort was made but if it can't explicitly inhibit suspending, we still want to perform the download & write operation.

[logiclrd]

Well shucks 🙂 There's no C.I., it seems. Hmm.

[tdewey-rpi]

Conceptually approved, but I can't merge it until I've put it through a round of testing.

This feature is sufficiently impactful that it should block 2.0 release.

I expect to come back to this PR on Thursday afternoon at the earliest.

[logiclrd]

a round of testing

Indeed :-) The main reason it was over a month between the issue and the PR is that I hate putting forth code I haven't tested myself. But, ultimately, I struggled with getting a development environment working and still wanted to contribute this feature, so here we are, a draft PR with no end-to-end testing yet. My apologies that that's the best I could muster.

[tdewey-rpi]

Broadly happy after initial review.

catch (...) patterns make me exceptionally nervous, but in this case the code you're guarding is relatively constrained, and not something where differentiated handling would offer an advantage, so I'm prepared to let it pass by.

[tdewey-rpi]

Tested, fixed up, and merged. If you want to see the fixups: 9d4a804

Thanks for a solid improvement!

[logiclrd]

Awesome, thanks so much 🙂