RFR: JDK-8136782: Introduce a build/configure wrapper (original) (raw)

Magnus Ihse Bursie magnus.ihse.bursie at oracle.com
Fri Dec 4 11:50:30 UTC 2015

Comments inline.

4 dec. 2015 kl. 10:58 skrev Erik Joelsson <erik.joelsson at oracle.com>:

Forgot the link http://cr.openjdk.java.net/~erikj/8136782/webrev.top.02/

On 2015-12-04 10:57, Erik Joelsson wrote: New webrev:

On 2015-12-03 22:01, Magnus Ihse Bursie wrote: On 2015-12-03 16:27, Erik Joelsson wrote: Hello,

This patch adds support for a new Oracle internal tool that will help us manage build configurations. I doubt it could be of any use to outside users, except perhaps to give a bit of insight into what versions of certain dependencies we use internally, as those will be specified in the jab-profiles.js configuration file. In addition to the JAB configuration files added here, I've also added a new configure option --with-default-make-target with which you can change the default make target for a configuration. A bug is fixed in the configure wrapper script where it's currently not propagating the return value of configure. Finally a new make command line variable is added, CONFNAME, which works like CONF, except it only matches exact configuration names. Bug: https://bugs.openjdk.java.net/browse/JDK-8136782 Webrev: http://cr.openjdk.java.net/~erikj/8136782/webrev.top.01/ Hi Erik, Nice to see progress on this new feature. :) I have a bunch of comments this time. Good! In basics.m4: The indentation looks weird, but maybe that's just an artifact of webrev? Fixed I'm a bit curious why this fix was needed. I thought one of the idea with BASICFIXUPPATH was to guarantee that we pointed to an existing directory? The documentation only says path, which I interpret as either a file or dir. If you needed to check something else than directories, maybe a different function should be used, like BASICFIXUPEXECUTABLE. I needed to (in closed) fix a file that wasn't executable. BASICFIXUPEXECUTABLE doesn't handle that. IMO, the change to support files as well as dirs is better than having to implement a complete new set of macros. Does --with-default-make-target really belong in build-performance? I'm not sure really where it belongs, though. It's more closely related to the stuff in source-dirs.m4, but the name of that file makes it too narrow. Maybe rename it to sources.m4 and put it there? I reacted to this too when I sorted through the changes I had made in the sandbox, but there was no really good fit. Perhaps basics.m4 is best? Moved it there

Looks ok.

In InitSupport.gmk: Maybe you should check so not both CONF and CONFNAME is used as well?

Also make/Help.gmk should be updated with CONFNAME, I think. Fixed # Create a SPEC definition. This will contain the path to one or more spec.gmk files. SPECS := (addsuffix /spec.gmk, (addprefix (builddir)/, (matchingconfs))) Fixed Actually, no. It will contain only one. In fact, maybe you should add a sanity check that matchingconfs contains exactly one configuration (unless it's empty). And, given that, the SPECS assignment can be simplified to just simple string concatenation. Right, fixed. In jab.sh: 25 # This script installs the jdk build wrapper tool into it's own local 26 # repository and puts wrapper scripts into this bin directory. "installs the jab tool" perhaps sounds better? Fixed the comment 35 if [ -f "~/.jab" ]; then 36 source ~/.jab Is this a local configuration file? Why not .config/jab/jab.conf instead? I'd like to keep my home dir clutter free, if possible. Sure, didn't know of any other standard, but .config seems to be pretty popular in my home dir. 97 getcmd{getcmd} getcmd{jaburl} > "${installedjabscript}.gz" 98 rm -f "${installedjabscript}" 99 gunzip "${installedjabscript}.gz" Maybe rm -f the "${installedjabscript}.gz" first as well, just to be on the safe side? Also, since network connectivity is the most likely cause of failure here, maybe explicitely test that "${installedjabscript}.gz" arrived and is non-zero, and otherwise print a clear error message stating the URL that we tried to access and say that it failed. And/or check the exit code from wget/curl? Added test for -s. Also, when installing jab, maybe you could be slightly more verbose? Like, "Downloading jab.sh.gz" and "Extracting jab.sh.gz", just to let the user know something is happening. And while jab.sh.gz might be likely to be small, on a slow line it might still be a noticable delay. In my experience, it's so small the delay is almost not noticeable, even when downloading from west coast US to Stockholm, but I can add some echos. Note that these will not look like other JAB output since the logging framework is not yet available before JAB has even started. 99 gunzip "${installedjabscript}.gz" Maybe we should verify that gunzip is available? Added. Do we need to run setupurl() even if not installing? I'm also not sure about the "Install url changed" thingy. If I've once downloaded jab using JABSERVER=foo jab.sh, then I should be able to leave JABSERVER=foo out on future invocations, methinks. But this logic looks like it's going to trip that up. I'm not sure for what reason. I did this intentionally. The reason is that stale downloads should not affect usage of the tool. The idea with JAB is to always use the latest version unless specifically told not to. The setupurl is needed to verify that the correct version of JAB is installed. Note that the url can change for a number of reasons, not just the base url part. Imagine someone changing this script to update the jabrevision, you would want that to be reflected next time the script is executed. Note that in the normal case, where you supply JABSERVER, that part of the URL is not saved in the data file, so leaving it out next time will not trigger a new download.

I see. I'm fine with this. Just one note, if you had jab_url=$jab_server/$data_string, maybe this should be more obvious. And it's a nice bit of DRY.

Finally, shell scripts with functions can be bit a hard to read. Maybe you could add some kind of delimiter after the installjab() function and before the main body of the file? Done Since the script is storing stuff in .jab, maybe you should add .jab to .hgignore? Forgot to bring that change over, added. In jab-profiles.js:

45 * The parameter 'input' is an object that optionally contains the 46 * some data. "the some" --> "some". * outputbasedir; "build", Perhaps colon? Fixed // List of free from labels describing aspects of this profile 115 * labels: "free form". But why not "tags" instead of "labels"? I'd say that in modern usage, "tag" means just "free form label". A discussion with Stefan, who is building the tool consuming this particular attribute, he preferred labels.

Hm. I think he's mistaken. :) But what the heck, who am I to say. :)

132 * // List of configure args to add when using this dependency, 133 * // defaults to input.get("<dependency-name", "installpath")

Seems a "--with-=" got lost there, somehow? Fixed 135 * // Name of environment variable to set when using this dependency 136 * environmentname: 137 * // Value of environment variable to set when using this dependency 138 * environmentvalue: 139 * // Value to add to the PATH variable when using this dependency 140 * environmentpath: This is set for "jab make" only, right? Maybe clarify that. Clarified 144 * // For certain dependencies where a legacy distribution mechanism is 145 * // already i place, the "javare" server layout is also supported "in place". Also, missing period. 150 * // buildnumber (optional), files and checksumfile is possible for "build number", "checksum file" * // aritfacts following the standard layout. "artifacts" * // For other files, use checksumpath and paths instead "checksum path" 62 * input.buildunixos 63 * input.buildunixcpu 66 * input.buildunixplatform What is buildunixcpu, etc? Sounds fully incomprehensible to me. :) Clarified

I still think this is not so good. What does build_unix_os contain? "Windows" or "Cygwin"? Is it used only on Windows? And why "unix"? Is the whole reason for this to be able to separate 32 and 64 bit cygwin?

In the build system, we have a notion of "os env", which is the same as "os" in most cases, but isn"cygwin" or "msys" on Windows. Maybe we should use the same terminology here, if we describe the same thing.

Except for the things I noted, your fixes look good.

/Magnus

/Erik /Magnus

More information about the build-dev mailing list