xparq / space_test Goto Github PK

While EXEC is a perfect counterpart of SH, and it also conveys exactly what it is, but only for devs...
I mean it's all for devs anyway, but for the overall style... It just reads shit. Like the examples in the README:

EXEC something
EXPECT some result

or

EXEC some command
EXEC other command
EXEC command --with-params

EXPECT "some result
....

Perhaps it just looks way too similar to EXPECT. Or is it just me?!

Maybe DO? Or just keep it RUN? Both are subpar for accuracy of meaning, but just taste better. Their genericity is also a plus, from the overall readability aspect (for high-level test authoring/maintenance), not just a minus (from the low-level technical test coding aspect).

• I think I'll just keep it RUN for now, and revisit this if really becomes untenable.

• Renaming run_cases is uncontroversial though. The mild crosstalk between RUN and run is kinda still annoying, but nothing that can't be lived with.

(Originally opened for Args (-> xparq/Args#30), but the test scripts now live here, so taking over...)

Now it thinks everything's a test case.

Oh, and of course run_cases run_cases is a nice infinite loop then! :)

Which makes the Args GHA test (-> #18) fail...

-> #10... So, ship it, or require it?

(Originally: xparq/Args#32)

# If listing more than one dependency, separate them with "," <-- or should it be ";"?!
Either way, eating up whitespace around the separator would also be a problem for the makefiles...

E.g.:

$ run_cases "subset*"

ERROR: Test case "subset*" not found!

$ run_cases "subset*/*"
realpath: C:/sz/prj/Space_Test/test/subset subdir/subtree case 1.case subset subdir/subtree case with build.case: No such file or directory

ERROR: Test case "subset*/*" not found!

Related ("dependent") issues (decisions that have simplified things a lot):

• TEST_DIR/run (and run.cmd) has been used as the sole reliable anchor for locating the TEST_DIR.. Where that script is, there's the test stuff. And from there it just also knows that _engine is also there, right underneath.

• The current logic is too brittle: it relies solely on $0 being in the _engine dir, which is not the case when run is being sourced from a script elsewhere, or if it was run via a symlink (at least on Windows; can't recall if sh resolves that symlink, hopefully not)!

• Now at least TEST_DIR can be set before calling run (which would then accept it, no questions asked).

• Support symlinks to _engine/run[.cmd] as anchors of the test dir!
  -> E.g.: https://github.com/cavo789/dos_batch_tips#getsymlinktargetpath

Well, following through from #11, let the config tell it.

Because the shebangs still had to remain #!/bin/sh -- but then it's a manifest inconsistency: sh doesn't support this syntax!

The implementations I encountered on Windows all still did tho, because many are just bashes in disguise.
So, all was well, until actually running in a real Linux shell, which took the shebangs seriously, and so the scripts broke...

That should also spare the extra subshell of SH echo!

I'm no shell programmer, you see. I've basically googled and manned and SXed all the .sh syntax & semantics together, so chances are I'm still missing a few glaring issues (even with such a simple language as ash).

So, it's only needed for bridging over to subprocesses, but not when sourcing scripts:

• spawned engine (sub)commands
• support for possibly launched user processes (anywhere) in the given context

I may have extra or missing exports basically anywhere...

(Originally: xparq/Args#33)

The .gitattributes settings supposedly already ensure that it's that way in the repo.
However, I still had core.autocrlf enabled locally, and either because of that or it was just never changed, (some of?) the scripts still had CRLF (in the work tree at least).

I recently dos2unixed them all, and also did a git add --renormalize (which didn't change anything then, but suppressed Git's really inexplicable "...will change to CRLF" warning); let's see what happens next time I check something out and feed it e.g. to WSL's bash...

Oh yeah, obviously? And then the CASE files, too, obviously, right? And then what about single-file test case scripts with natural text as their names? Should they also be "Startup with defaults.CASE"? I mean not that "Startup with defaults.case" would look any better, but capitals in the extension is a HIGHLY unusual habit, and may be offputting to some.

OK then... Make it configurable! :) See also #12 then!

E.g. these are some of the vars exported for a build, showing what I mean:

export TEST_NAME
export TEST_DIR
export CASE
export TEST_CASE_DIR
...

And... While we are at it:

• Stop differentiating between the top level and the stuff below!
  This is a tree, and it should just be a purely recursive structure!
  • Or did I mean it specifically for the build support?! Fork the build-specific parts of this issue into another then!

...so e.g. a missing test exe problem would be more obvious to recognize.

Not to mention the current feature of also accepting Git's sh!... :-o

The issue is that busybox versions obviously differ vastly, so relying on "found" versions in a host env. is a hit-and-miss.
OTOH, bundling a version for every possible supported env. seems to be both futile and restrictive. (Or is it? First, it's only really necessary on Windows, and second, there seems to be a (32-bit?) "canonical" version at Ron's page anyway!...)

NOTE: now (-> #22) it's actually also being downloaded on-the-fly, too, so that's also an uncontrolled version dependency issue!

So, the tightening, and/or various levels of shell (mostly BB) version control part is still relevant!

(Originally opened as xparq/Args#35)

This is to help debugging individual cases without entering the test case dir manually just for executing a local exe there, or even just changing the last command too much...

In this mode the EXPECT phase will likely be off, so a warning should be printed.

Note: the params would be appended to every RUN/SH statement of the case.

That should also make it easier to "integrate" (copy...) into other projects, esp. upgrading existing ones (better separation from test data (artifacts) and the "engine")!

See e.g. the README case with 3 echo lines, each with NL: expected even if there's no NL at the end of the EXPECT text!

Ehheh, OK, fine, fine... I meant "spaces in explicitly named test case names", it seems... :)

Anyway, it was an easy fix: just replacing $* with $@ in the main runner iterator loop, as per https://linux.die.net/man/1/ash:

 @     Expands to the positional parameters, starting from one. When the expansion occurs within
       double-quotes, each positional parameter expands as a separate argument. If there are
       no positional parameters, the expansion of @ generates zero arguments, even when @ is 
       double-quoted. 
       What this basically means, for example, is if $1 is ''abc'' and $2 is ''def ghi'', then "$@" expands 
       to the two arguments:
       "abc" "def ghi"

Woahh... OK, so it had nothing to do with EXPECT being empty, actually! :)

It was a problem of trying to RUN echo, which must be run via SHELL.

-> #25

You see, that's exactly what I was talking about in #9...

Well, there are two Git shs... Oh, wailt, three even, actually!... :-o And the "wrong" one is on the PATH here.

1. On the PATH, unusable here: "C:\Program Files\Git\usr\bin\sh.exe"
2. Not on the PATH, but the one we need: "C:\Program Files\Git\bin\sh.exe"
3. And I have 0 clue what exactly this one does differently, but unusable too: "C:\Program Files\Git\git-bash.exe"

Test:

set "GITROOT=C:\Program Files\Git"
where sh.exe
rem        -> C:\Program Files\Git\usr\bin\sh.exe
where find
rem        -> C:\Windows\System32\find.exe
rem        -> C:\Program Files\Git\usr\bin\find.exe
@echo.
@echo This will use Windows's find.exe:
sh.exe -c "which find; find . -name '*.*'"
@echo.
@echo This will use Git's find:
"%GITROOT%\bin\sh.exe" -c "which find; find . -name '*.*'"
@echo.
@echo WTF is this trying to do?
"%GITROOT%\git-bash.exe" -c "which find; find . -name '*.*'"

Also, critically: don't forget -c (the error message is horribly misleading)!

>"...\sh.exe" find -- version
find: find: cannot execute binary file

So:

sh -c "find --version"

run_cases "TC*"
- ERROR: Test case "TC dir with no suffix TC dir with no suffix and no script TC dir with suffix, but no script.case" not found!

Unquoted run_cases TC* is fine.

But still allow overriding it with any local Makefile* just like in the other test dirs.

(Originally: xparq/Args#20)

Regex support in the EXPECT clauses could also do, but that won't address e.g. standalone EXPECT files.

E.g., it was also a "wish item" in the Args README case, which is, indeed, has a split run-script vs. EXPECT case setup:

# CLEAN s/[^ ]+.+\.exe/\.\.\.\.exe/

-> #20

$ run_cases *bui*
Riding "...\prj\Space_Test\test\_engine/busybox" sh...
"...\prj\Space_Test\test\_engine\busybox" sh -c ". build.sh \"test-demo\""
realpath: .../prj/Space_Test/test/*bui*: Invalid argument

ERROR: Test case "*bui*" not found!

-------- THERE HAVE BEEN FAILED CASES!

$ run_cases *exe*
Riding "...\prj\Space_Test\test\_engine/busybox" sh...
gnumake: 'test-demo.exe' is up to date.

ERROR: Test case "general.exe" not found!


ERROR: Test case "test-demo.exe" not found!

-------- THERE HAVE BEEN FAILED CASES!

I mean better... It's not obvious to first-time visitors that there's anything interesting here at all!

This should fix the missing BB error in the Args GHA test run...

Esp. this confusing one:

SH "echo Dir list:; ls -r"
SH echo command 3
EXPECT "Dir list:
Hi from the test case dir!
CASE
command 3
"

(And some other things below that.)

Yeah, well, sounds scarier than it really is (it's a relic I know of, and it's harmless), but still kinda facepalm...
(The thing, BTW, that makes the difference (but shouldn't), is having or not having the $TEST_CASE_FILE_EXT in the TC dir's name.)

-> #42

WSL:

.../prj/Space_Test/test$ ./run_cases
./run_cases: 213: [: gcc: unexpected operator
./run_cases: 280: [: gnumake: unexpected operator
./run_cases: 326: /mnt/c/sz/prj/Space_Test/test/_engine/init_once.sh: gnumake: not found

• It's probably some bash vs. ash incompatibility regressions, mostly...
• The direct gnumake dependency in functions.sh is gross... It's not even called "gnumake" by default. ;)
  • Either ship with one (err... for every platform?!... :)) ), or detect before calling!
  • Note: using BB-w32's built-in make would help even less here: a) it's only available there, on Windows, b) it's very limited, and c) there's likely an NMAKE also available there anyway...

Currently:

>bash -c ./run_cases general
./run_cases: 12: [: unexpected operator
./run_cases: 48: [: 0: unexpected operator
-------- THERE HAVE BEEN FAILED CASES!  

Is even [ "x" == "y" ] a bashism, with that ==?!

-> shellcheck -s bash ...

Notes copied from a dup. (#48):

WSL:

.../prj/Space_Test/test$ ./run_cases
./run_cases: 213: [: gcc: unexpected operator
./run_cases: 280: [: gnumake: unexpected operator
./run_cases: 326: /mnt/c/sz/prj/Space_Test/test/_engine/init_once.sh: gnumake: not found

• It's probably some bash vs. ash incompatibility regressions, mostly...
• The direct gnumake dependency in functions.sh is gross... It's not even called "gnumake" by default. ;)
  • Either ship with one (err... for every platform?!... :)) ), or detect before calling!
  • Note: using BB-w32's built-in make would help even less here: a) it's only available there, on Windows, b) it's very limited, and c) there's likely an NMAKE also available there anyway...
• Add a Linux GitHub Action, too!

Well, the current 1-depth initial MVP is gonna be too tight a fit soon, obviously...

Too much incompatibilities...
E.g. the makefiles can't just glue the build script path uniformly to $_SH_, because under Git sh it's /c/shit/..., while BB.exe has no problem with the normal C:/... paths... :-/

OTOH, if it's too explicit, or supports BB alone, then it may not just run fine in a native sh env as it should!...
But we could still stick to BB on Windows at least... (E.g. in cases where there's no $SHELL yet? Nope: $SHELL doesn't help with the Git sh vs BB_win32 path discrepancy!)

As an option, as that's entirely (intrusively) C++-specific!
-> e.g. https://github.com/xparq/regenv/blob/master/src/FindListPart.cpp

Updated it there to 0.08.
Fails at the GHA right now, for missing BB... -> #22

The test runner has been passing $MAKE for a while now, so the actual build command could be selected accordingly.