Delete dmd/src/dmd

[RazvanN7]

After modifying all the scrips to use dmd/generated/os/build/model/dmd instead of dmd/src/dmd, we can delete the dmd/src/dmd binary.

[wilzbach]

Could you please remove the src/dmd.conf target as well? Thanks!

[RazvanN7]

I assume you were referring to $G/dmd.conf

[RazvanN7]

CC @braddr , @CyberShadow , @MartinNowak

[CyberShadow]

In win32.mak:

G = $(GEN)\$(OS)$(MODEL)

OS is not defined anywhere.

[CyberShadow]

Also, the directory structure is very different on Windows and POSIX.

[CyberShadow]

Noting that to reliably use the new paths, one needs to either glob, or either invoke or reimplement osmodel.mak in their code. This is a regression in complexity. Perhaps it would be an improvement to allow specifying the desired output path of the DMD binary as a Makefile parameter.

[RazvanN7]

@CyberShadow From what I could deduce, the directory hierarchies are very different when it comes to posix vs windows builds. For example, if you look at phobos : the linux version has a generated/os/build/model hierarchiy, while the windows version employs a totally different hierarchy (there is a src folder where all the sources are kept - dmd, druntime, phobos - alltogheter and a bin directory where libraries and binaries are generated). Seeing this inconsistency, I thought that the build chains are considered to be different and as a result the hierarchies are different. I will add in win32.mak the os variable which will be set to windows and the build variable which will be set to release, but I don't think that will make any difference.

[CyberShadow]

Yep, all true. But if we are to define new hierarchies, might as well make them consistent with themselves, right?

[CyberShadow]

Addressed issues in comments

How come @MartinNowak is the committer (or author?) of that commit? And was that meant to be an amended commit?

[RazvanN7]

@CyberShadow It was intended to be an amended commit, but I have no idea why Martin appeared to coauthor the commit. I rebased and tried to commit again but now it looks like dlang-bot is the coauthor of this commit

[CyberShadow]

I think you might have rebased/amended too far and included the previous merge commit on master.

[CyberShadow]

I'd suggest trying to amend with --reset-author.

[RazvanN7]

@CyberShadow Looking through my commit logs I saw that the first commit on this branch accidentally included --amend, but --reset-author solved the case. Thanks

[CyberShadow]

OK, though you still should rebase on master. Right now your last commit has two parents, and is a merge commit between master and the tip of PR #6572.

[RazvanN7]

@andralex @CyberShadow @wilzbach I think this is ready to be merged. Next stop: ddmd -> dmd

[wilzbach]

@RazvanN7 I think sth went wrong during your rebase - have a look at the changes tab

[RazvanN7]

@wilzbach It seems that circleci needs a dmd.conf in generated

[andralex]

Well things are failing for the time being. How do we kickstart this?

[RazvanN7]

@MartinNowak After this PR : dlang/druntime#1779 one would expect that the autotester uses dmd in generated/os/release/model. Why isn't that happening?

[RazvanN7]

@wilzbach I had to keep the dmd.conf target otherwise circleci and travis would fail.

[andralex]

@RazvanN7 please take a look at the auto-tester (https://auto-tester.puremagic.com/pull-history.ghtml?projectid=1&repoid=1&pullid=6595), seems to still need src/dmd. E.g. from Darwin_64_64:

../dmd/src/dmd -conf= -c -o- -Isrc -Iimport -Hfimport/core/sync/barrier.di src/core/sync/barrier.d
make: ../dmd/src/dmd: No such file or directory

[andralex]

@MartinNowak @braddr @CyberShadow @wilzbach summoning the wizards for a puzzling matter with the autotester.

I see the autotester fails on e.g. FreeBSD_32 with:

../dmd/src/dmd -conf= -c -o- -Isrc -Iimport -Hfimport/core/sync/barrier.di src/core/sync/barrier.d
gmake: ../dmd/src/dmd: Command not found
posix.mak:146: recipe for target 'import/core/sync/barrier.di' failed

This looks trivial, enough, except https://github.com/dlang/druntime/blob/master/posix.mak has already been changed to use generated/ instead of src/ for locating the dmd binary, and in fact even the line given in the error message (146) is off by one (should be 147). That suggests the druntime makefile run by the autotester is older than the one in master.

What could cause this?

[rainers]

What could cause this?

The auto-tester passes it explicitly on the command line: https://github.com/braddr/at-client/blob/master/src/do_build_druntime.sh#L15. Same for phobos.

[andralex]

@rainers thanks! braddr/at-client#1

[andralex]

@wilzbach is this related to your recent work?

[andralex]

@RazvanN7 please rebase so we keep this in good shape until we sort it out

[RazvanN7]

I think that we can close this since it is out of sync and @wilzbach 's PR does the same thing and much more.