[llvm] [Lit] Introduce a new IO interface for in-process built-ins (PR #194664)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 28 09:08:41 PDT 2026
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-testing-tools
Author: Benjamin Stott (BStott6)
<details>
<summary>Changes</summary>
This PR is part of a series of patches upgrading Lit's in-process built-ins to be able to run with piped input/output and full redirection support, and to allow custom in-process builtns to be provided via the Lit config. The remaining patches to Lit's test runner can be found [here](https://github.com/BStott6/llvm-project/compare/lit-inproc-builtins). This is part of the [Lit daemonized testing project](https://discourse.llvm.org/t/rfc-reducing-process-creation-overhead-in-llvm-regression-tests/88612/9).
Currently, in-process built-ins have no way to take data on `stdin`, and must return their `stdout` and `stderr` as string objects. This is insufficient for the daemonization project, as we need to be able to provide input to the in-process built-ins calling the daemon tools, and we need to be able to handle binary output from the daemons. This introduces a new IO interface for in-process built-ins, with binary IO objects for `stdin`, `stdout` and `stderr` which may be open file objects or in-memory streams (`BytesIO`). The constructor for the `InprocBuiltinIO` object is intended to mirror how IO streams are provided to `subprocess.Popen`: the arguments can be open binary file objects, `subprocess.PIPE`/`subprocess.STDOUT` sentinels or None.
The existing in-process built-ins are updated to use the new interface. The code in `TestRunner.py` that calls the in-process built-ins is updated accordingly. The tests verifying that in-process built-ins fail as expected when `STDIN` is redirected to come from a file are updated to check that these pass instead, as this is now supported behaviour. The `glob-echo` test is also updated to pass now, as we now call `expand_glob_expressions` for in-process built-ins.
---
Full diff: https://github.com/llvm/llvm-project/pull/194664.diff
4 Files Affected:
- (modified) llvm/utils/lit/lit/InprocBuiltins.py (+175-76)
- (modified) llvm/utils/lit/lit/TestRunner.py (+38-16)
- (modified) llvm/utils/lit/tests/shtest-glob.py (+1-2)
- (modified) llvm/utils/lit/tests/shtest-shell.py (+5-32)
``````````diff
diff --git a/llvm/utils/lit/lit/InprocBuiltins.py b/llvm/utils/lit/lit/InprocBuiltins.py
index 34a4d7eae1e88..9bc43f4d30143 100644
--- a/llvm/utils/lit/lit/InprocBuiltins.py
+++ b/llvm/utils/lit/lit/InprocBuiltins.py
@@ -5,77 +5,158 @@
import shutil
import stat
import subprocess
-from io import BytesIO, StringIO
+from dataclasses import dataclass
+from io import BytesIO
+from typing import Any, Callable, Dict, List, Optional
import lit.util
+from lit.ShCommands import Command
from lit.ShellEnvironment import (
InternalShellError,
- ShellCommandResult,
- expand_glob_expressions,
+ ShellEnvironment,
kIsWindows,
- processRedirects,
updateEnv,
)
-def executeBuiltinCd(cmd, shenv):
+class InprocBuiltinIO:
+ """
+ Holds IO streams for an inproc builtin invocation.
+
+ NB: If stderr is redirected to be the same stream as stdout, then
+ `stder == stdout` is True.
+ """
+
+ stdin: Any
+ stdout: Any
+ stderr: Any
+
+ def __init__(self, stdin, stdout, stderr):
+ """
+ Configure the IO streams for an in-process built-in command. This
+ constructor is designed to mirror how IO streams are configured in
+ `subprocess.Popen`.
+
+ Each of the arguments to this constructor may be:
+ - A file object open in binary mode.
+ - `subprocess.PIPE` or `subprocess.STDOUT` sentinels.
+ - None.
+ """
+
+ # If stderr is redirected to stdout, we make sure to use the same
+ # stream for both so that the order of output is preserved.
+ stderr_redirected_to_stdout = (
+ stdout == subprocess.PIPE and stderr == subprocess.STDOUT
+ )
+
+ # Replace sentinel values with in-memory streams.
+ if stdin == subprocess.PIPE or stdin is None:
+ self.stdin = BytesIO()
+ else:
+ self.stdin = stdin
+
+ if stdout == subprocess.PIPE or stdout is None:
+ self.stdout = BytesIO()
+ else:
+ self.stdout = stdout
+
+ if stderr_redirected_to_stdout:
+ # Make sure stderr and stdout are directed to the same stream.
+ self.stderr = self.stdout
+ elif stderr == subprocess.PIPE or stderr is None:
+ self.stderr = BytesIO()
+ else:
+ self.stderr = stderr
+
+
+InprocBuiltinExecuteFn = Callable[
+ [Command, List[str], ShellEnvironment, InprocBuiltinIO],
+ int,
+]
+"""
+Function called by an in-process built-in command.
+Parameters:
+ - `cmd`: The command itself.
+ - `args`: glob-expanded list of arguments (including argv[0] as the program
+ name).
+ - `shenv`: The shell environment.
+ - `io`: Holds the input and output streams for the invocation. These are
+ binary IO objects (files open in binary mode, BytesIO).
+
+The return value is the exit code.
+"""
+
+
+ at dataclass
+class InprocBuiltin:
+ """
+ Represents a command that is run as an in-process built-in.
+ """
+
+ execute: InprocBuiltinExecuteFn
+ """
+ Function called to execute the in-process built-in.
+ """
+
+ fallback: Optional[str] = None
+
+
+def executeBuiltinCd(
+ cmd: Command, args: List[str], shenv: ShellEnvironment, io: InprocBuiltinIO
+) -> int:
"""executeBuiltinCd - Change the current directory."""
- if len(cmd.args) != 2:
+ if len(args) != 2:
raise InternalShellError(cmd, "'cd' supports only one argument")
# Update the cwd in the parent environment.
- shenv.change_dir(cmd.args[1])
+ shenv.change_dir(args[1])
# The cd builtin always succeeds. If the directory does not exist, the
# following Popen calls will fail instead.
- return ShellCommandResult(cmd, "", "", 0, False)
+ return 0
-def executeBuiltinPushd(cmd, shenv):
+def executeBuiltinPushd(
+ cmd: Command, args: List[str], shenv: ShellEnvironment, io: InprocBuiltinIO
+) -> int:
"""executeBuiltinPushd - Change the current dir and save the old."""
- if len(cmd.args) != 2:
+ if len(args) != 2:
raise InternalShellError(cmd, "'pushd' supports only one argument")
shenv.dirStack.append(shenv.cwd)
- shenv.change_dir(cmd.args[1])
- return ShellCommandResult(cmd, "", "", 0, False)
+ shenv.change_dir(args[1])
+ return 0
-def executeBuiltinPopd(cmd, shenv):
+def executeBuiltinPopd(
+ cmd: Command, args: List[str], shenv: ShellEnvironment, io: InprocBuiltinIO
+) -> int:
"""executeBuiltinPopd - Restore a previously saved working directory."""
- if len(cmd.args) != 1:
+ if len(args) != 1:
raise InternalShellError(cmd, "'popd' does not support arguments")
if not shenv.dirStack:
raise InternalShellError(cmd, "popd: directory stack empty")
shenv.cwd = shenv.dirStack.pop()
- return ShellCommandResult(cmd, "", "", 0, False)
+ return 0
-def executeBuiltinExport(cmd, shenv):
+def executeBuiltinExport(
+ cmd: Command, args: List[str], shenv: ShellEnvironment, io: InprocBuiltinIO
+) -> int:
"""executeBuiltinExport - Set an environment variable."""
- if len(cmd.args) != 2:
+ if len(args) != 2:
raise InternalShellError(cmd, "'export' supports only one argument")
- updateEnv(shenv, cmd.args)
- return ShellCommandResult(cmd, "", "", 0, False)
+ updateEnv(shenv, args)
+ return 0
-def executeBuiltinEcho(cmd, shenv):
+def executeBuiltinEcho(
+ cmd: Command, args: List[str], shenv: ShellEnvironment, io: InprocBuiltinIO
+) -> int:
"""Interpret a redirected echo or @echo command"""
opened_files = []
- stdin, stdout, stderr = processRedirects(cmd, subprocess.PIPE, shenv, opened_files)
- if stdin != subprocess.PIPE or stderr != subprocess.PIPE:
- raise InternalShellError(
- cmd, f"stdin and stderr redirects not supported for {cmd.args[0]}"
- )
-
- # Some tests have un-redirected echo commands to help debug test failures.
- # Buffer our output and return it to the caller.
- is_redirected = True
- if stdout == subprocess.PIPE:
- is_redirected = False
- stdout = BytesIO()
# Implement echo flags. We only support -e and -n, and not yet in
# combination. We have to ignore unknown flags, because `echo "-D FOO"`
# prints the dash.
- args = cmd.args[1:]
+ args = args[1:]
interpret_escapes = False
write_newline = True
while len(args) >= 1 and args[0] in ("-e", "-n"):
@@ -94,30 +175,24 @@ def maybeUnescape(arg):
if args:
for arg in args[:-1]:
- stdout.write(maybeUnescape(arg).encode())
- stdout.write(b" ")
- stdout.write(maybeUnescape(args[-1]).encode())
+ io.stdout.write(maybeUnescape(arg).encode())
+ io.stdout.write(" ".encode())
+ io.stdout.write(maybeUnescape(args[-1]).encode())
if write_newline:
- stdout.write("\n".encode())
+ io.stdout.write("\n".encode())
for name, mode, f, path in opened_files:
f.close()
- output = (
- ""
- if is_redirected
- # TODO(BStott) remove decode once new interface for in-process builtin
- # IO is introduced.
- else stdout.getvalue().decode(encoding="utf8", errors="replace")
- )
- return ShellCommandResult(cmd, output, "", 0, False)
+ return 0
-def executeBuiltinMkdir(cmd, cmd_shenv):
+def executeBuiltinMkdir(
+ cmd: Command, args: List[str], cmd_shenv: ShellEnvironment, io: InprocBuiltinIO
+):
"""executeBuiltinMkdir - Create new directories."""
- args = expand_glob_expressions(cmd.args, cmd_shenv.cwd)[1:]
try:
- opts, args = getopt.gnu_getopt(args, "p")
+ opts, args = getopt.gnu_getopt(args[1:], "p")
except getopt.GetoptError as err:
raise InternalShellError(cmd, "Unsupported: 'mkdir': %s" % str(err))
@@ -131,7 +206,6 @@ def executeBuiltinMkdir(cmd, cmd_shenv):
if len(args) == 0:
raise InternalShellError(cmd, "Error: 'mkdir' is missing an operand")
- stderr = StringIO()
exitCode = 0
for dir in args:
dir = pathlib.Path(dir)
@@ -144,16 +218,19 @@ def executeBuiltinMkdir(cmd, cmd_shenv):
try:
dir.mkdir(exist_ok=True)
except OSError as err:
- stderr.write("Error: 'mkdir' command failed, %s\n" % str(err))
+ io.stderr.write(
+ ("Error: 'mkdir' command failed, %s\n" % str(err)).encode()
+ )
exitCode = 1
- return ShellCommandResult(cmd, "", stderr.getvalue(), exitCode, False)
+ return exitCode
-def executeBuiltinRm(cmd, cmd_shenv):
+def executeBuiltinRm(
+ cmd: Command, args: List[str], cmd_shenv: ShellEnvironment, io: InprocBuiltinIO
+):
"""executeBuiltinRm - Removes (deletes) files or directories."""
- args = expand_glob_expressions(cmd.args, cmd_shenv.cwd)[1:]
try:
- opts, args = getopt.gnu_getopt(args, "frR", ["--recursive"])
+ opts, args = getopt.gnu_getopt(args[1:], "frR", ["--recursive"])
except getopt.GetoptError as err:
raise InternalShellError(cmd, "Unsupported: 'rm': %s" % str(err))
@@ -176,7 +253,6 @@ def on_rm_error(func, path, exc_info):
os.chmod(path, stat.S_IMODE(os.stat(path).st_mode) | stat.S_IWRITE)
os.remove(path)
- stderr = StringIO()
exitCode = 0
for path in args:
cwd = cmd_shenv.cwd
@@ -189,7 +265,7 @@ def on_rm_error(func, path, exc_info):
os.remove(path)
elif os.path.isdir(path):
if not recursive:
- stderr.write("Error: %s is a directory\n" % path)
+ io.stderr.write(("Error: %s is a directory\n" % path).encode())
exitCode = 1
if platform.system() == "Windows":
# NOTE: use ctypes to access `SHFileOperationsW` on Windows to
@@ -255,26 +331,28 @@ class SHFILEOPSTRUCTW(Structure):
os.chmod(path, stat.S_IMODE(os.stat(path).st_mode) | stat.S_IWRITE)
os.remove(path)
except OSError as err:
- stderr.write("Error: 'rm' command failed, %s" % str(err))
+ io.stderr.write(("Error: 'rm' command failed, %s" % str(err)).encode())
exitCode = 1
- return ShellCommandResult(cmd, "", stderr.getvalue(), exitCode, False)
+ return exitCode
-def executeBuiltinUmask(cmd, shenv):
+def executeBuiltinUmask(
+ cmd: Command, args: List[str], shenv: ShellEnvironment, io: InprocBuiltinIO
+):
"""executeBuiltinUmask - Change the current umask."""
if os.name != "posix":
raise InternalShellError(cmd, "'umask' not supported on this system")
- if len(cmd.args) != 2:
+ if len(args) != 2:
raise InternalShellError(cmd, "'umask' supports only one argument")
try:
# Update the umask in the parent environment.
- shenv.umask = int(cmd.args[1], 8)
+ shenv.umask = int(args[1], 8)
except ValueError as err:
raise InternalShellError(cmd, "Error: 'umask': %s" % str(err))
- return ShellCommandResult(cmd, "", "", 0, False)
+ return 0
-def executeBuiltinUlimit(cmd, shenv):
+def executeBuiltinUlimit(cmd: Command, args: List[str], shenv, io: InprocBuiltinIO):
"""executeBuiltinUlimit - Change the current limits."""
try:
# Try importing the resource module (available on POSIX systems) and
@@ -282,34 +360,55 @@ def executeBuiltinUlimit(cmd, shenv):
import resource
except ImportError:
raise InternalShellError(cmd, "'ulimit' not supported on this system")
- if len(cmd.args) != 3:
+ if len(args) != 3:
raise InternalShellError(cmd, "'ulimit' requires two arguments")
try:
- if cmd.args[2] == "unlimited":
+ if args[2] == "unlimited":
new_limit = resource.RLIM_INFINITY
else:
- new_limit = int(cmd.args[2])
+ new_limit = int(args[2])
except ValueError as err:
raise InternalShellError(cmd, "Error: 'ulimit': %s" % str(err))
- if cmd.args[1] == "-v":
+ if args[1] == "-v":
if new_limit != resource.RLIM_INFINITY:
new_limit = new_limit * 1024
shenv.ulimit["RLIMIT_AS"] = new_limit
- elif cmd.args[1] == "-n":
+ elif args[1] == "-n":
shenv.ulimit["RLIMIT_NOFILE"] = new_limit
- elif cmd.args[1] == "-s":
+ elif args[1] == "-s":
if new_limit != resource.RLIM_INFINITY:
new_limit = new_limit * 1024
shenv.ulimit["RLIMIT_STACK"] = new_limit
- elif cmd.args[1] == "-f":
+ elif args[1] == "-f":
shenv.ulimit["RLIMIT_FSIZE"] = new_limit
else:
- raise InternalShellError(
- cmd, "'ulimit' does not support option: %s" % cmd.args[1]
- )
- return ShellCommandResult(cmd, "", "", 0, False)
+ raise InternalShellError(cmd, "'ulimit' does not support option: %s" % args[1])
+ return 0
-def executeBuiltinColon(cmd, cmd_shenv):
+def executeBuiltinColon(
+ cmd: Command, args: List[str], cmd_shenv: ShellEnvironment, io: InprocBuiltinIO
+):
"""executeBuiltinColon - Discard arguments and exit with status 0."""
- return ShellCommandResult(cmd, "", "", 0, False)
+ return 0
+
+
+def get_default_inproc_builtins() -> Dict[str, InprocBuiltin]:
+ """
+ get_default_inproc_builtins - Returns the map of command names to Lit's
+ in-process built-in implementations.
+ """
+
+ return {
+ "@echo": InprocBuiltin(executeBuiltinEcho, None),
+ "cd": InprocBuiltin(executeBuiltinCd, None),
+ "export": InprocBuiltin(executeBuiltinExport, None),
+ "echo": InprocBuiltin(executeBuiltinEcho, None),
+ "mkdir": InprocBuiltin(executeBuiltinMkdir, None),
+ "popd": InprocBuiltin(executeBuiltinPopd, None),
+ "pushd": InprocBuiltin(executeBuiltinPushd, None),
+ "rm": InprocBuiltin(executeBuiltinRm, None),
+ "ulimit": InprocBuiltin(executeBuiltinUlimit, None),
+ "umask": InprocBuiltin(executeBuiltinUmask, None),
+ ":": InprocBuiltin(executeBuiltinColon, None),
+ }
diff --git a/llvm/utils/lit/lit/TestRunner.py b/llvm/utils/lit/lit/TestRunner.py
index ee91342e969d9..09502c8e9e4c1 100644
--- a/llvm/utils/lit/lit/TestRunner.py
+++ b/llvm/utils/lit/lit/TestRunner.py
@@ -11,11 +11,11 @@
import threading
import traceback
-import lit.InprocBuiltins as InprocBuiltins
import lit.ShUtil as ShUtil
import lit.Test as Test
import lit.util
from lit.BooleanExpression import BooleanExpression
+from lit.InprocBuiltins import InprocBuiltinIO, get_default_inproc_builtins
from lit.ShCommands import Command
from lit.ShellEnvironment import (
InternalShellError,
@@ -254,19 +254,7 @@ def _executeShCmd(cmd, shenv, results, timeoutHelper):
builtin_commands_dir = os.path.join(
os.path.dirname(os.path.abspath(__file__)), "builtin_commands"
)
- inproc_builtins = {
- "cd": InprocBuiltins.executeBuiltinCd,
- "export": InprocBuiltins.executeBuiltinExport,
- "echo": InprocBuiltins.executeBuiltinEcho,
- "@echo": InprocBuiltins.executeBuiltinEcho,
- "mkdir": InprocBuiltins.executeBuiltinMkdir,
- "popd": InprocBuiltins.executeBuiltinPopd,
- "pushd": InprocBuiltins.executeBuiltinPushd,
- "rm": InprocBuiltins.executeBuiltinRm,
- "ulimit": InprocBuiltins.executeBuiltinUlimit,
- "umask": InprocBuiltins.executeBuiltinUmask,
- ":": InprocBuiltins.executeBuiltinColon,
- }
+ inproc_builtins = get_default_inproc_builtins()
# To avoid deadlock, we use a single stderr stream for piped
# output. This is null until we have seen some output using
# stderr.
@@ -354,9 +342,43 @@ def _executeShCmd(cmd, shenv, results, timeoutHelper):
j,
"Unsupported: '{}' cannot be part" " of a pipeline".format(args[0]),
)
- result = inproc_builtin(Command(args, j.redirects), cmd_shenv)
+
+ stdin, stdout, stderr = processRedirects(
+ j, subprocess.PIPE, shenv, opened_files
+ )
+
+ builtin_io = InprocBuiltinIO(stdin, stdout, stderr)
+
+ args = expand_glob_expressions(args, cmd_shenv.cwd)
+
+ exit_code = inproc_builtin.execute(
+ Command(args, j.redirects), args, cmd_shenv, builtin_io
+ )
+
+ builtin_io.stdout.flush()
+ builtin_io.stderr.flush()
+
if not_count % 2:
- result.exitCode = int(not result.exitCode)
+ exit_code = int(not exit_code)
+
+ # Gather output from the streams.
+ out = ""
+ if stdout == subprocess.PIPE:
+ builtin_io.stdout.seek(0)
+ out = builtin_io.stdout.read().decode("utf8", "replace")
+
+ err = ""
+ if stderr == subprocess.PIPE:
+ builtin_io.stderr.seek(0)
+ err = builtin_io.stderr.read().decode("utf8", "replace")
+
+ result = ShellCommandResult(
+ j,
+ out,
+ err,
+ exit_code,
+ False,
+ )
result.command.args = j.args
results.append(result)
return result.exitCode
diff --git a/llvm/utils/lit/tests/shtest-glob.py b/llvm/utils/lit/tests/shtest-glob.py
index 95a53e4dd2223..190745f8133d4 100644
--- a/llvm/utils/lit/tests/shtest-glob.py
+++ b/llvm/utils/lit/tests/shtest-glob.py
@@ -4,8 +4,7 @@
# RUN: | FileCheck -dump-input=fail -match-full-lines --implicit-check-not=Error: %s
# END.
-# CHECK: UNRESOLVED: shtest-glob :: glob-echo.txt ({{[^)]*}})
-# CHECK: AttributeError: 'GlobItem' object has no attribute 'encode'
+# CHECK: PASS: shtest-glob :: glob-echo.txt ({{[^)]*}})
# CHECK: FAIL: shtest-glob :: glob-mkdir.txt ({{[^)]*}})
# CHECK: # | Error: 'mkdir' command failed, {{.*}}example_file1.input'
diff --git a/llvm/utils/lit/tests/shtest-shell.py b/llvm/utils/lit/tests/shtest-shell.py
index 498f6bb0adc11..fa5d63b5e9e29 100644
--- a/llvm/utils/lit/tests/shtest-shell.py
+++ b/llvm/utils/lit/tests/shtest-shell.py
@@ -497,37 +497,10 @@
# CHECK-NEXT: # error: command failed with exit status: 1
# CHECK: ***
-# CHECK: FAIL: shtest-shell :: echo-at-redirect-stderr.txt
-# CHECK: *** TEST 'shtest-shell :: echo-at-redirect-stderr.txt' FAILED ***
-# CHECK: @echo 2> {{.*}}
-# CHECK: # executed command: @echo
-# CHECK: # .---command stderr{{-*}}
-# CHECK: # | stdin and stderr redirects not supported for @echo
-# CHECK: error: command failed with exit status:
-
-# CHECK: FAIL: shtest-shell :: echo-at-redirect-stdin.txt
-# CHECK: *** TEST 'shtest-shell :: echo-at-redirect-stdin.txt' FAILED ***
-# CHECK: @echo < {{.*}}
-# CHECK: # executed command: @echo
-# CHECK: # .---command stderr{{-*}}
-# CHECK: # | stdin and stderr redirects not supported for @echo
-# CHECK: error: command failed with exit status:
-
-# CHECK: FAIL: shtest-shell :: echo-redirect-stderr.txt
-# CHECK: *** TEST 'shtest-shell :: echo-redirect-stderr.txt' FAILED ***
-# CHECK: echo 2> {{.*}}
-# CHECK: # executed command: echo
-# CHECK: # .---command stderr{{-*}}
-# CHECK: # | stdin and stderr redirects not supported for echo
-# CHECK: error: command failed with exit status:
-
-# CHECK: FAIL: shtest-shell :: echo-redirect-stdin.txt
-# CHECK: *** TEST 'shtest-shell :: echo-redirect-stdin.txt' FAILED ***
-# CHECK: echo < {{.*}}
-# CHECK: # executed command: echo
-# CHECK: # .---command stderr{{-*}}
-# CHECK: # | stdin and stderr redirects not supported for echo
-# CHECK: error: command failed with exit status:
+# CHECK: PASS: shtest-shell :: echo-at-redirect-stderr.txt
+# CHECK: PASS: shtest-shell :: echo-at-redirect-stdin.txt
+# CHECK: PASS: shtest-shell :: echo-redirect-stderr.txt
+# CHECK: PASS: shtest-shell :: echo-redirect-stdin.txt
# CHECK: FAIL: shtest-shell :: error-0.txt
# CHECK: *** TEST 'shtest-shell :: error-0.txt' FAILED ***
@@ -634,4 +607,4 @@
# CHECK: PASS: shtest-shell :: valid-shell.txt
# CHECK: Unresolved Tests (1)
-# CHECK: Failed Tests (37)
+# CHECK: Failed Tests (33)
``````````
</details>
https://github.com/llvm/llvm-project/pull/194664
More information about the llvm-commits
mailing list