Download release file instead of cloning repo in go-template (continued from #179) #180
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Rather than use a bare string, we capture the output in an array.
This provides a better error than a raw `[[ -f ... ]]`.
Don't know why the test wasn't originally written using `TEST_GO_ROOTDIR`. As intended, it exposed some missing quotes from variables in `go-template` introduced in #179. I've removed several instances of the following comment, as I'm not sure it applies anymore. If it does, I'll make sure to restore it as needed: # Use `.*/scripts/go-script-bash` to account for the fact that `git clone` on # MSYS2 will output `C:/Users/<user>/AppData/Local/Temp/` in place of `/tmp`.
Also wraps a couple of lines that were previously longer than 80 columns.
These aren't necessary, as `stub_program_in_path` will update `PATH` automatically.
By using a process substitution to provide input to `tar`, we eliminate a subshell and the need for `set -o pipefail`.
Seems cleaner and easier to follow to give both operations descriptive names and allow them to reside at the top level.
This prevents new processes from spawning just to detect whether a program is present. The new `create_forwarding_script` helper in `tests/template.bats` ensured that the tests continued to pass before and after this change. `create_forwarding_script` will move to `lib/bats/helpers` in a future commit.
In addition to being more specific, the errors are now printed to standard error.
Defining these variables at the top level keeps all the environment variables defined in one consistent location, and makes them available to suite names as well.
Now all of the lines are back down to 80 characters or less.
This had been hardcoded to the actual repo value in #179. The `create_fake_tarball_if_not_using_real_url` helper function and `TEST_ARCHIVE_URL` environment variable enable the test to run isolated from the network. Also tweaks `GO_SCRIPTS_DIR` to remove unnecessary braces.
Got rid of `GO_SCRIPT_BASH_VERSION_NUMBER`, since it was only used in this function. Elided some of the `if` conditions into a single `if/elif` chain. Made sure that the `mkdir -p` tries to create the full parent of `GO_SCRIPT_BASH_CORE_DIR`, not just `GO_SCRIPTS_DIR`.
Attempts to be flexible as the `./go get file` command. The `run_with_download_program` test helper creates a script that runs `cat` on a local tarball to simulate downloads using each tool.
The `download uses ...` tests all restrict the avaiable programs to a very small subset via forwarding scripts. Apparently GNU tar, unlike BSD tar, calls out to the `gzip` program rather than linking in the library. So this change adds a forwarding script for `gzip` to fix this particular breakage: * https://travis-ci.org/mbland/go-script-bash/jobs/268538634
As I'd already done with the 'fail to download a nonexistent version' test case, I've updated the 'fail to download a nonexistent repo' test case to use a series of `assert_output_matches` assertions rather than checking the literal output with `assert_failure`. This avoids the problem whereby tool+platform-specific error messages can cause the tests to fail: * https://travis-ci.org/mbland/go-script-bash/jobs/268538634
This is in preparation for using `cp` instead of `wget` for `file://` urls, since `wget` only supports HTTP, HTTPS, and FTP.
This is because `wget` only supports HTTP, HTTPS, and FTP, as I discovered when running the tests on Alpine Linux without cURL installed. Also, the BusyBox version of `tar` that comes with Alpine doesn't work out of the box; the GNU version must be installed via `apk add tar`. Not sure it's worth adding a check to `go-template` or not.
This fixes test failures for the "fail to find download program uses git clone" and the "fail to find tar uses git clone" test cases on MSYS2 and Cygwin, due to the fact that helper programs couldn't be found with the `PATH`. This resulted in errors such as: C:/msys64/usr/lib/git-core/git-upload-pack.exe: error while loading shared libraries: ?: cannot open shared object file: No such file or directory
Cygwin and MSYS2 environments both contain the `cygpath` utility for translating between Unix and native Windows file paths. `cygpath -m` is just what's needed when passing file paths and URLs to native binary programs such as `curl` and `git`. This commit adds the `windows_native_path` helper function, which I'll eventually extract into a `lib/` module at some point. It also adds logic to ensure `GO_SCRIPT_BASH_DOWNLOAD_URL` begins with a protocol, adds a test case to validate this, removes redundant comments and assertions from several `git clone` test cases, and updates the `create_forwarding_script` helper to only generate a script when a program is present on the system.
As it turns out, regular MSYS2 and Cygwin do not work with with file paths modified with `cygpath -m`. This change tries to ensure only Git for Windows is affected by checking the `EXEPATH` variable.
Turns out I made a mistake in the previous commit in adapting the conditional to use `EXEPATH` instead of `command -v cygpath`, as the tests were still broken on regular MSYS2 and Cygwin. This change resolves the breakage and continues to work on Git for Windows.
Turns out this is the most reliable and robust way, as `EXEPATH` won't always be set or set with 'Git' as its final component.
Turns out the 'fail to find tar uses git clone' test would fail on systems that don't have `curl` installed, since `create_forwarding_script` now only creates a script if a program is present. This fixes the breakage by trying to create a forwarding script for all three programs; at least one should be guaranteed to be present.
|
Wow, it literally took me all day to get things smoothed out across platforms. Specifically, I came to realize how particular the Git for Windows environment is, even compared to plain MSYS2 and Cygwin. But now I'm armed with insight from that experience (including my discovery of the Thanks for the contribution, @elpaquete! Though since it's almost 9pm and I haven't had dinner yet, I think I'll cut v1.6.0 tomorrow. 🙂 |
This was referenced Aug 27, 2017
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Closes #179.
I've taken @elpaquete's fine work from #179 and refactored it a bit to:
master(after I remembered to push the v1.5.0 commit just today—d'oh!)wgetandfetchas well.I still need to test this on all the platforms (I think I might need to add tweaks for MSYS2), but I hope to have it in by tomorrow and cut v1.6.0 right away.
@elpaquete Note that I found a way to get rid of
curl | tarandset -o pipefail, namely by using a command substitution and checking that the expected output is present.