This change addresses issues reported by the `staticcheck` tool and a
couple reported by the `golint` tool. It also adds missing issue links
to a couple of entries in the release notes. This change deliberately
does not address these warnings since it is unclear whether the project
owner would prefer to suppress or address them:
pkg/store/cmd.go:8:2: should not use dot imports (ST1001)
pkg/store/db_store.go:10:2: should not use dot imports (ST1001)
pkg/store/dir.go:8:2: should not use dot imports (ST1001)
Introduce `builtin:buildinfo` and `builtin:version` vars. This also
changes the `-buildinfo -json` implementation to use the standard
encoding/json package rather than handcrafting the JSON string.
This also fixes three incorrect spellings of "overridden". Normally I would
do those in a separate change but since there are only three instances,
one of which one was legitimately part of this change, I decided to bundle
the other two.
In particular, the code for retrieving file paths needed by the runtime is
simplified a lot, and the paths only needed for the storage daemon is cleanly
separated from other paths.
There was a recent failure of the test that checks peach is
nondeterministic: https://github.com/elves/elvish/runs/2926941131
The failure can be reproduced reliably when setting GOMAXPROCS=1.
However, GitHub Action's Windows runner has two CPU cores, which means
GOMAXPROCS should be 2. Presumably, during that particular one, one of
the cores is occupied by another process, so Go had to run the
goroutines in order on one CPU core.
Go's time.Sleep yields the current goroutine and allow other scheduled
goroutines to run on the same thread. As a result, adding a random
jitter guarantees nondeterminism in execution order, even if only one
CPU core is available to execute threads.
Also run the nondeterminism check in an infinite loop.
This fixes the *Segment.Concat and RConcat methods to handle the new
number types. It also adds more tests of concatenation with the "Text"
and "Segment" types with number types.
Related #1340
For some reason this wrong sequencing was only discovered on FreeBSD, and the
test reliably pass on macOS (likely differences in scheduling algorithms).
Also make the bug a bit easier to reproduce on FreeBSD by using an external
command on the reader side. Builtin commands are too quick to execute.
Two different strategies are used:
- For commands that don't actually depend on the editor (but happens to live
inside the edit: module), their tests are changed to use the evaltest
framework.
- For commands that do depend on the editor, a new helper
testThatOutputErrorIsBubbled is used to test them.
Also fix a bug where edit:complete-getopt does not correctly bubble the
exception thrown from callbacks, and where matchers do not bubble output errors.
The dependency is injected via a field in shell.Program.
Also add an alternative entrypoint, cmd/nodaemon/elvish, which does not include
the daemon server or client.
Also replace (*Frame).OutputChan with (*Frame).ValueOutput, which returns a
small interface for writing to the value output that is also aware when the
reader is gone.
A test in navigation_mode.go verifies that a certain key has no effect. It does
that by verifying that the buffer has not changed, which results in a race
condition: the verification might happen after the key is handled, in which
case the test works. But the verification may also happen before the key is
actually handled, in which case the key handling code has not run and thus has
no coverage.
Fix this by queueing another key that would result in an unrelated buffer
change, thus indirectly verifying that the previous key had no effect.
This also replaces the slightly awkward "arguments here" reason with
"argument count" as the "what" for a typical errs.ArityMismatch
exception. It also reformats most of the constructors so that the "what"
is on the same line. This makes `grep errs.ArityMismatch **.go` more
useful as a result.
Also simplify the logic of AltGr handling. It seems that AltGr and
AltGr+Shift don't actually combine with any other modifiers, so we can
test whether filteredMod is leftCtrl|rightAlt or
leftCtrl|rightAlt|shift, and leave the Mod field unpopulated in the
return value.
This is an unusual change in that it exists solely to exercise generated
code and thus increase test coverage. On the one hand, if we never care
about the string value of type `RedirMode` then it should be removed
from the stringified types. On the other hand, verifying a redirection
is correctly represented in the output of the parse.pprint...() family
of functions is useful given the importance of I/O redirections in shells
like Elvish.
This change makes feeding output to commands which handle NUL terminated
"lines" (e.g., `fzf -read0` or `xargs -0`) extremely fast compared to
using an explicit Elvish loop that does `print $val"\x00"`. Similarly for
handling input from commands that produce NUL terminated "lines" (e.g.,
`find . -print0`) compared to an Elvish loop using `read-upto "\x00"`.
Resolves#1070
Related #1053
It is in theory better implemented as a StructMap because it is a transparent
data type. However, the reflection-based algorithm for StructMap will create a
"kind" field for it, and we don't want to remove the custom kind of Pipe yet, so
this has to be a PseudoStructMap now.
In light of the preceding change to address staticcheck lint warnings I
ran this:
grep 'errors\.New("' **.go |
sed -e 's/^.*errors\.New("\([^"]*\)".*/\1/' |
sort | uniq -c | sort -n
Which caused me to notice two errors associated with the builtin `sleep`
command have multiple definitions. It is questionable whether the negative
duration error is justified. I think it should be replaced by the invalid
duration error but decided not to do that in order to limit the scope of
this change.
Fix two issues found by `staticcheck -tests=false ./...`:
pkg/eval/go_fn.go:155:5: var errNoOptions is unused (U1000)
pkg/persistent/hashmap/hashmap.go:321:2: only the first constant in this group has an explicit type (SA9004)
I was looking at "TODO" comments that could be eliminated and noticed the
one preceding the sole use of `isDrive()`. Which caused me to notice its
implementation is broken. The `s[1] < 'z'` test should be `s[0] < 'z'`
I noticed that testutil.MustPipe was not covered by any unit tests. This
converts the handful of places that should use it to do so. This changes
the coverage of pkg/testutil/must.go from 61.5% to 73.1%. There isn't
any way to increase that further without explicitly testing the panic
paths.
This is useful when piping the output into a program like `fzf` and more
efficient than dealing with the map. This takes 180.32 ms on my system
(best of five runs):
```
time { edit:command-history &dedup &newest-first | each [hist-entry]{ print $hist-entry[cmd]"\000" } >/dev/null }
```
This option reduces the time to 156.12 ms:
```
time { edit:command-history &dedup &newest-first &cmd-only | each [cmd]{ print $cmd"\000" } >/dev/null }
```
It's not a huge difference but is 13% faster, and should be considerably
faster when combined with a hypothetical `to-lines &null` option I intend
to implement.
Related #1053
This implements `&dedup` and `&newest-first` options for
`edit:command-history`. This makes it noticably cheaper to feed unique
command history into external commands like `fzf`.
Related #1053Fixes#568
While working on issue #568 to add a `&dedup` option to the
`edit:command-history` command I noticed that the `store.Cmd.Cmds()`
method should be removed.
Rather than having specialized commands make a `file:pipe` object
indexable so we can use the generic `file:close` command. This does not
address existing problems; such as builtins not failing when writing to
a `file:pipe` object if the read-end is closed.
Related #1316
Creating symlinks requires a special permission on Windows, which the
use may not have.
Also remove the support for symlinks in testutil.ApplyDir. Because of
this issue on Windows, any test that requires creating symlinks needs to
be broken out to an individual test case, meaning that symlinks can't be
created alongside other files, so there is less benefit of keeping it
inside the same Dir structure.
The `edit:after-command` hooks are called with a single argument:
a pseudo-map with these keys:
"command": the command line that was run
"duration": the execution duration in seconds
"error": any error that occurred ($nil if no error occurred)
The `edit:command-duration` variable is the elapsed seconds (as a
float64) of the most recently run interactive command.
Resolves#1029
* Emphasize that it's incomplete.
* Remove the TODO for documenting the default binding; it can be
inspected in an Elvish instance, it is intentionally left
undocumented.
* Remove `MatchesRegexp` from eval/vals since it is not part of the
Elvish value protocol.
* Simplify implementation of value matching in `eval/evaltest`.
* Documentation wording tweaks.
POSIX shells have a `-i` short option to force interactive mode. Some
programs, such as the `script` command, assume that option is recognized
by the shell it spawns. There isn't any reason elvish shouldn't silently
ignore that option.
Resolves#1292
These tests exercise the arity checking logic, which is implemented
generically for every builtin function and does not need to be tested
individually.
Prclose and pwclose with Deprecation marked in specific file. Give
feedback regarding the test cases.
File module now has prclose and pwclose
Wrote test cases for prclose and pwclose, need better test cases
Deprecation for prclose and pwclose marked in respective files.
Fix `//elv:fn` to the correct `//elvdoc:fn` form. Add a little more
information about the "command" mode of the Elvish interactive editor.
Related #971
Using a read-only variable as the target of an `except` clause should
highlight just the var name rather than the entire `try...except...`
statement.
Resolves#1258
When attempting to update a read-only var return an error struct rather
than a simple string (i.e., a Go `error` type). This makes it possible
to include the var name in the error message. This builds on commit
a33ecb2d that highlights the offending var name in the stack trace but
does not include the var name in the error message. With this change the
error message includes the offending var name.
Related #255
The "-count" flag causes the test functions to be invoked multiple times. Since
some tests mutate their test data or some global state, they will fail when
invoked with a count larger than 1.
This commit changes *some* of such tests to either avoid mutating the test data,
or resetting the global state. Some such tests still remain and will be fixed later.
A new type, tk.Bindings is defined for bindings. It is always called with the
widget currently handling the event, in addition to the event being handled.
* Move all widget types to a new pkg/cli/tk.
* Move Mode{Line Prompt} to a new pkg/cli/mode.
* Move all packages in pkg/cli/addons to pkg/cli/mode.
* Remove WriteListing.
* Remove StdTTY.
* Rename GetCodeBuffer to CodeBuffer.
* Make the Override function remove the override when given a negative width,
instead of panicking.
* Make all functions concurrency-safe.
* Improve test coverage to 100%.
This addresses #1234.
A Query is a sequence of compound expressions or pairs, like the arguments
and options in a command form. It is not used by any other node type, and is
intended to be a top-level node for representing a small DSL for filtering data,
such as the filter for history mode or location mode.
This addresses #1051.
* Rebrand NEXT-RELEASE.md to the release notes for 0.15.0, and publish it.
* Bump default deprecation level to 15.
* Add download links in website/get/prelude.md.
The mechanism is now documented in PACKAGING.md.
Also refactor tools/buildall.sh to make it easy to make reproducible builds, and
fix tools/cirrus-deploy.sh.
Previously, to avoid showing deprecation warnings for the next release when the
user is running on HEAD, a boolean CLI flag -show-deprecations is introduced,
and is set to false in the master branch. The idea is that release branches will
have this default to true, so people running released versions will see
deprecations.
However, this means that people running on HEAD will never see any deprecations
unless they use this CLI flag, which is not ideal. This commit replaces the flag
bool -show-deprecations with a numerical -deprecation-level flag, which requests
deprecations that are active as of release 0.X to be shown. The default value of
this flag will be the minor version number of the *last* release, so that people
running HEAD will see as many deprecation warnings as people running the last
release would. This number will be bumped just before releases.
* Introduce the `path:` module
This is based on https://github.com/elves/elvish/pull/1084 by @kolbycrouch
submitted five months ago. It addresses all of the feedback on that
change and includes other documentation and unit test improvements. It
also includes a couple of extensions to the original P.R., such as a
`path:is-abs` command.
I decided to resurrect that change because I want better support for
filesystem path manipulation so that users can replace non-portable
external commands such as `realpath` and `find` with Elvish builtins. This
is a baby step towards that goal.
Related #849
* Add a `path:is-regular` command
This adds a `path:is-regular` command. This is for symmetry with the
`path:is-dir` command and the glob `[type:regular]` modifier.
It also adds support for symlinks in the `testutil.Applydir` function
and change the path unit test to use it.
* Rename path:real to path:eval-symlinks
The Elvish daemon should not inherit file descriptors open on the tty.
Make the daemon's stdin open on the null device and its stdout/stderr
open on a predictable file.
Fixes#1191
