Repeated terminfo db opens

[rbtcollins]

Problem
rustup is opening the terminfo db repeatedly, which is wasteful.

Steps
Run rustup under strace / process monitor; debug builds are useful to get good stacks :).

3:54:41.7794556 PM	rustup.exe	972	CreateFile	C:\Users\robertc\.terminfo	NAME NOT FOUND	Desired Access: Read Attributes, Synchronize, Disposition: Open, Options: Synchronous IO Non-Alert, Attributes: n/a, ShareMode: Read, Write, Delete, AllocationSize: n/a
3:54:41.7795392 PM	rustup.exe	972	CreateFile	C:\etc\terminfo	PATH NOT FOUND	Desired Access: Read Attributes, Synchronize, Disposition: Open, Options: Synchronous IO Non-Alert, Attributes: n/a, ShareMode: Read, Write, Delete, AllocationSize: n/a
3:54:41.7795954 PM	rustup.exe	972	CreateFile	C:\lib\terminfo	PATH NOT FOUND	Desired Access: Read Attributes, Synchronize, Disposition: Open, Options: Synchronous IO Non-Alert, Attributes: n/a, ShareMode: Read, Write, Delete, AllocationSize: n/a
3:54:41.7796514 PM	rustup.exe	972	CreateFile	C:\usr\share\terminfo	PATH NOT FOUND	Desired Access: Read Attributes, Synchronize, Disposition: Open, Options: Synchronous IO Non-Alert, Attributes: n/a, ShareMode: Read, Write, Delete, AllocationSize: n/a
3:54:41.7797046 PM	rustup.exe	972	CreateFile	C:\boot\system\data\terminfo	PATH NOT FOUND	Desired Access: Read Attributes, Synchronize, Disposition: Open, Options: Synchronous IO Non-Alert, Attributes: n/a, ShareMode: Read, Write, Delete, AllocationSize: n/a

This shows up multiple times

A sample stack fragment

14	rustup.exe	std::sys::windows::fs::File::open + 0x1e7, /rustc/2aa4c46cfdd726e97360c2734835aa3515e8c858\/src\libstd\sys\windows\fs.rs(256)	0x7ff66d20fa77	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
15	rustup.exe	std::sys::windows::fs::stat + 0x57, /rustc/2aa4c46cfdd726e97360c2734835aa3515e8c858\/src\libstd\sys\windows\fs.rs(676)	0x7ff66d210ae7	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
16	rustup.exe	std::fs::metadata<std::path::PathBuf*> + 0x4e, /rustc/2aa4c46cfdd726e97360c2734835aa3515e8c858\src\libstd\fs.rs(867)	0x7ff66c4691fe	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
17	rustup.exe	term::terminfo::searcher::get_dbpath_for_term + 0xb49, C:\Users\robertc\.cargo\registry\src\github.com-1ecc6299db9ec823\term-0.5.2\src\terminfo\searcher.rs(73)	0x7ff66c445dc9	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
18	rustup.exe	term::terminfo::TermInfo::from_name + 0x65, C:\Users\robertc\.cargo\registry\src\github.com-1ecc6299db9ec823\term-0.5.2\src\terminfo\mod.rs(93)	0x7ff66c457315	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
19	rustup.exe	term::terminfo::TermInfo::from_env + 0x178, C:\Users\robertc\.cargo\registry\src\github.com-1ecc6299db9ec823\term-0.5.2\src\terminfo\mod.rs(85)	0x7ff66c457168	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
20	rustup.exe	term::terminfo::TerminfoTerminal<std::io::stdio::Stderr>::new<std::io::stdio::Stderr> + 0x44, C:\Users\robertc\.cargo\registry\src\github.com-1ecc6299db9ec823\term-0.5.2\src\terminfo\mod.rs(384)	0x7ff66c45a394	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
21	rustup.exe	term::stderr + 0x20, C:\Users\robertc\.cargo\registry\src\github.com-1ecc6299db9ec823\term-0.5.2\src\lib.rs(1)	0x7ff66c4462e0	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe
22	rustup.exe	rustup_init::term2::stderr + 0x9, C:\Users\robertc\Documents\src\rustup.rs\src\cli\term2.rs(56)	0x7ff66c41c289	C:\Users\robertc\Documents\src\rustup.rs\target\debug\rustup.exe

Possible Solution(s)

The term library does no caching of the environment; it looks likes complete minimisation requires retaining the terminfo db and use term::terminfo::TerminfoTerminal new_with_terminfo rather than the stdout and stderr helpers.

A probably fine intermediate step is to open stdout and stderr just once each and manage those resources.

Notes

This is in git as of 373448b
Output of rustup --version:
Output of rustup show:

[kinnison]

How many times do we currently do this? The majority of the uses of the term instances seem to be cached and passed around. The stdout/stderr helpers instantiate a TerminfoTerminal which does seem to cache the content. Basically only calls to term2::std*() create new instances which, while called moderately commonly, doesn't seem to be done in any performance critical areas (e.g. the terminal for the download tracker is cached)

[rbtcollins]

cli/log.rs seems to be particularly evident when I sample the stack traces

[kinnison]

I concur that the log functions reopen the terminal, we could lazy_static it perhaps.

[kinnison]

Looks like term is unmaintained and we should move away from it anyway -- #1818 is related to that.

[rbtcollins]

Hah. Well I have a thing that makes it all better, just playing rust golf, combined with road tripping. Will try to put it up today.

…

On Mon, 29 Apr 2019, 19:57 Daniel Silverstone, ***@***.***> wrote: Looks like term is unmaintained and we should move away from it anyway -- #1818 <#1818> is related to that. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1800 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AADZ7XQTOW3GESBSB6JZESDPS2S5LANCNFSM4HIAHZYA> .

[rbtcollins]

Ok, have looked at termcolor; (ug the spelling :( ). It's no different in terms of primary public API - in that it still returns a thing we have to manage ourselves to avoid repeated requerying of the underlying context; it does appear to do a somewhat lighterweight job of that - by virtue of not honouring terminfo at all.

[rbtcollins]

#1820

[rbtcollins]

Fixed with 1820.