-
Notifications
You must be signed in to change notification settings - Fork 29
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
silver init doesn't recognize login shells (e.g. "-bash", "-su") #67
Labels
bug
Something isn't working
enhancement
New feature or request
good first issue
Good for newcomers
medium
Medium difficulty
Comments
thecaralice
added
good first issue
Good for newcomers
enhancement
New feature or request
bug
Something isn't working
medium
Medium difficulty
labels
Nov 14, 2020
zackw
added a commit
to zackw/silver
that referenced
this issue
Nov 24, 2020
All silver subcommands look up the parent process’s name, and use that, verbatim, to identify the shell in use. This has several problems: - On Windows, the parent process’s name may or may not end with “.exe”; there was special case logic to handle “powershell.exe” versus “powershell” but not “bash.exe” versus “bash”. Relatedly, the parent process’s name ought to be interpreted case- insensitively on Windows; BASH.EXE is the same as bash.exe. - On Unix, there is a notion of a _login_ shell (invoked directly by login(1) for console logins, or sshd(1) for remote logins); these are identified by a leading dash in their process name, *not* by a special option. There was no code to handle this at all (bug reujab#67). - sysinfo is slow, because it reads a whole bunch of files in /proc whether it’s going to need the data within or not. Some but not all of this overhead can be eliminated by using `System::new_with_specifics(RefreshKind::new())` instead of `System::new_all()`, but even then the wall-clock time for `silver init` drops from 0.021s to 0.002s (release build) if sysinfo is not invoked at all. This patch addresses all these problems, as follows: - There is a new command line option `--shell` which may be used to explicitly set the shell in use. Also, support for the `SILVER_SHELL` environment variable (used for this purpose in 1.1 and previous) is restored. sysinfo is only consulted if neither the command line option, nor the environment variable, is available. - No matter where we get the process name from, it is lowercased, a trailing “.exe” is removed, a leading dash is removed, and any leading directory names are also removed (bug reujab#16). - The initialization scripts set `SILVER_SHELL`, so sysinfo will be consulted at most once per session, not once per command. I also reorganized the code in the Command::Init branch of main for readability’s sake, and made the “unknown shell” error message not be a panic. (Panics are for _bugs_, not for errors with a cause outside the program.)
zackw
added a commit
to zackw/silver
that referenced
this issue
Nov 24, 2020
All silver subcommands look up the parent process’s name, and use that, verbatim, to identify the shell in use. This has several problems: - On Windows, the parent process’s name may or may not end with “.exe”; there was special case logic to handle “powershell.exe” versus “powershell” but not “bash.exe” versus “bash”. Relatedly, the parent process’s name ought to be interpreted case- insensitively on Windows; BASH.EXE is the same as bash.exe. - On Unix, there is a notion of a _login_ shell (invoked directly by login(1) for console logins, or sshd(1) for remote logins); these are identified by a leading dash in their process name, *not* by a special option. There was no code to handle this at all (bug reujab#67). - sysinfo is slow, because it reads a whole bunch of files in /proc whether it’s going to need the data within or not. Some but not all of this overhead can be eliminated by using `System::new_with_specifics(RefreshKind::new())` instead of `System::new_all()`, but even then the wall-clock time for `silver init` drops from 0.021s to 0.002s (release build) if sysinfo is not invoked at all. This patch addresses all these problems, as follows: - There is a new command line option `--shell` which may be used to explicitly set the shell in use. Also, support for the `SILVER_SHELL` environment variable (used for this purpose in 1.1 and previous) is restored. sysinfo is only consulted if neither the command line option, nor the environment variable, is available. - No matter where we get the process name from, it is lowercased, a trailing “.exe” is removed, a leading dash is removed, and any leading directory names are also removed (bug reujab#16). - The initialization scripts set `SILVER_SHELL`, so sysinfo will be consulted at most once per session, not once per command. I also reorganized the code in the Command::Init branch of main for readability’s sake, and made the “unknown shell” error message not be a panic. (Panics are for _bugs_, not for errors with a cause outside the program.)
zackw
added a commit
to zackw/silver
that referenced
this issue
Nov 24, 2020
All silver subcommands look up the parent process’s name, and use that, verbatim, to identify the shell in use. This has several problems: - On Windows, the parent process’s name may or may not end with “.exe”; there was special case logic to handle “powershell.exe” versus “powershell” but not “bash.exe” versus “bash”. Relatedly, the parent process’s name ought to be interpreted case- insensitively on Windows; BASH.EXE is the same as bash.exe. - On Unix, there is a notion of a _login_ shell (invoked directly by login(1) for console logins, or sshd(1) for remote logins); these are identified by a leading dash in their process name, *not* by a special option. There was no code to handle this at all (bug reujab#67). - sysinfo is slow, because it reads a whole bunch of files in /proc whether it’s going to need the data within or not. Some but not all of this overhead can be eliminated by using `System::new_with_specifics(RefreshKind::new())` instead of `System::new_all()`, but even then the wall-clock time for `silver init` drops from 0.021s to 0.002s (release build) if sysinfo is not invoked at all. This patch addresses all these problems, as follows: - There is a new command line option `--shell` which may be used to explicitly set the shell in use. Also, support for the `SILVER_SHELL` environment variable (used for this purpose in 1.1 and previous) is restored. sysinfo is only consulted if neither the command line option, nor the environment variable, is available. - No matter where we get the process name from, it is lowercased, a trailing “.exe” is removed, a leading dash is removed, and any leading directory names are also removed (bug reujab#16). - The initialization scripts set `SILVER_SHELL`, so sysinfo will be consulted at most once per session, not once per command. I also reorganized the code in the Command::Init branch of main for readability’s sake, and made the “unknown shell” error message not be a panic. (Panics are for _bugs_, not for errors with a cause outside the program.)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Labels
bug
Something isn't working
enhancement
New feature or request
good first issue
Good for newcomers
medium
Medium difficulty
silver init
looks up the name of its parent process (or, in older versions, at the value of theSILVER_SHELL
environment variable, which you're expected to set to$0
, which comes to the same thing) and uses that to decide what type of shell is running.If the parent process is a login shell (e.g. invoked by
/usr/bin/login
when you log in on a text console, or bysshd
when you log in remotely) then the parent process will have a dash prepended to its name (e.g.-bash
instead ofbash
). silver does not recognize this convention and will print a panic message to the effect of "unknown shell: -bash; supported shells: ..."silver init
should trim a leading dash from theshell
variable before matching on it.[N.B. with some implementations of
su
it's possible for you to get "-su" as the name of the login shell. Rather than trying to deduce the actual identity of the shell from other clues I would suggest restoring theSILVER_SHELL
environment variable and treating it as a way for the user to override silver's idea of the shell in use; when it's set, don't muck around with sysinfo at all.]The text was updated successfully, but these errors were encountered: