fix Issue 519 - Invariant not called from autogenerated class/struct ctor/dtor

[WalterBright]

fix https://issues.dlang.org/show_bug.cgi?id=519

Will follow up with change to spec.

[WalterBright]

Documentation fix: dlang/dlang.org#699

[WalterBright]

Phew! finally passing.

[yebblies]

Please don't add disabled code.

[WalterBright]

I prefer to leave it in until the controversy is over.

[yebblies]

Do you mean leave it in the pull request or leave it in the compiler's source?

[WalterBright]

yes and yes until the controversy is over.

[dnadlinger]

"Controversy is over" – where does the actual discussion take place? The last related reply in the bugzilla issue was in 2012.

[WalterBright]

Part of my solution is to change the spec to say the invariant won't run for trivial construction. Since obviously several people expect it to, I await their response.

[MartinNowak]

Sounds like we should have that discussion before making the change.

[MartinNowak]

http://forum.dlang.org/post/m4dk8s$2ktm$1@digitalmars.com

[WalterBright]

Sigh. I cannot reproduce these failures in std.datetime:

../dmd/src/dmd -O -w -L/co -c -unittest -ofunittest3b.obj std\datetime.d
std\datetime.d(17339): Error: function std.datetime.Interval!(Date).Interval.end need 'this' to access member end
std\datetime.d(17372):        called from here: end()
std\datetime.d(18694): Error: template instance std.datetime.Interval!(Date) error instantiating

Anyone know what is going on?

[WalterBright]

Figured it out. Needs this phobos pull:

dlang/phobos#2689

[yebblies]

Any idea why it worked before?

[WalterBright]

I don't know how datetime works. It's got too many layers to it :-(

[dnadlinger]

Typo in comment.

[MartinNowak]

I don't know how datetime works. It's got too many layers to it :-(

Can we actually please figure out why you had to rewrite legit code.
You can use dustmite to reduce test-cases.

The problem only appeared on Win which is quite weird when you look at your phobos pull.
Also now there is a new Win32 issue (https://auto-tester.puremagic.com/show-run.ghtml?projectid=1&runid=1256128&isPull=true).

[WalterBright]

Also now there is a new Win32 issue (https://auto-tester.puremagic.com/show-run.ghtml?projectid=1&runid=1256128&isPull=true).

This can be reproduced by adding the line:

public pure nothrow @safe @nogc ~this() { }

to struct Date and running with the master dmd. It's some other problem.

[MartinNowak]

This error looks a bit like the pull messed up trySemantic, just a guess though.

[WalterBright]

Added it as: https://issues.dlang.org/show_bug.cgi?id=13719

[MartinNowak]

Any objections against this change in general?

[MartinNowak]

@WalterBright it would be fairly interesting to get your opinion on this.
http://forum.dlang.org/post/m4eu6v$trq$1@digitalmars.com

[MartinNowak]

I'm uneasy with this. Why would you need to check the invariant before destroying structs/classes that don't have a destructor? No code could ever use that value again.
Before this change one could leave back such a value in any state, now I might have to do additional cleanup before it is destroyed.

[MartinNowak]

Ping, @WalterBright what do you think?

[WalterBright]

Why would you need to check the invariant before destroying structs/classes that don't have a destructor?

Same as always, to check for bugs. Live objects should always be in a valid state, even if they are headed for the gallows.

[WalterBright]

what do you think?

I think this should be pulled.

[WalterBright]

it would be fairly interesting to get your opinion on this.

I replied in the n.g.

[andralex]

@WalterBright we should pull this. Rebase?

[WalterBright]

rebased for ddmd

[andralex]

@WalterBright @MartinNowak status?

[andralex]

Why would you need to check the invariant before destroying structs/classes that don't have a destructor? No code could ever use that value again.

Walter is right - object should be valid for destruction

[andralex]

@WalterBright need rebase

[dlang-bot]

Fix	Bugzilla	Description
✓	519	Invariant not called from autogenerated class/struct constructor/destructor

[andralex]

legit errors in autotester

[andralex]

ping @WalterBright

[ibuclaw]

Rebased patch in #7536.