www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Mihaela Chirea - SAOC 2020 Milestone 1 Update 4 - Improving DMD as a

reply Mihaela Chirea <chireamihaela99 gmail.com> writes:
Hello!

Here is an update of the progress made during the last week of 
Milestone 1 as part of #SAOC2020:
- I mostly worked on the token location offset tests[1]. This 
included studying the lexer, parser and tokens in more detail to 
make sure all the tokens are getting the right offset.
- Updated the list of locations[2] to show the nodes missing the 
start location more clearly. It currently contains most nodes in 
attrib.d and DebugCondition and VersionCondition from cond.d.

For next week:
- Add the start location to the nodes that lack this information, 
starting with the ones from cond.d

These tasks were mentioned in the previous update[3] and will be 
done during the second Milestone:
- Bring all DMD as a library additions hidden behind a version 
under DMDLIB
- Add the token size

The initial milestones can be found here[4]. I will post the 
updated list as a reply to that thread and use that link in 
future updates.

[1]: https://github.com/dlang/dmd/pull/11823
[2]: 
https://drive.google.com/file/d/1lb-eP9Pai713tggud6iuzA1cIetbK4zc/view?usp=sharing
[3]: 
https://forum.dlang.org/post/vshahvctwdhrmrjssool forum.dlang.org
[4]: 
https://forum.dlang.org/post/ztbjfjebxtysmugfzkmz forum.dlang.org
Oct 18 2020
next sibling parent reply Mihaela Chirea <chireamihaela99 gmail.com> writes:
While going through the token types I also noticed that TOK.arrow 
seems to be unused. Is there a reason for keeping it? Same for 
int128 and uns128 but I thought those might be implemented in the 
future. Is TOK.arrow in a similar situation?
Oct 19 2020
parent reply =?UTF-8?Q?S=c3=b6nke_Ludwig?= <sludwig outerproduct.org> writes:
Am 19.10.2020 um 17:59 schrieb Mihaela Chirea:

 While going through the token types I also noticed that TOK.arrow seems 
 to be unused. Is there a reason for keeping it? Same for int128 and 
 uns128 but I thought those might be implemented in the future. Is 
 TOK.arrow in a similar situation?
 


I think the only reason is to give a helpful error message in case 
someone ports code from C/C++ or writes in that fashion. This is what 
DMD outputs:

Error: use . for member lookup, not ->
Oct 19 2020
parent reply RazvanN <razvan.nitu1305 gmail.com> writes:
On Monday, 19 October 2020 at 17:37:46 UTC, Sönke Ludwig wrote:

 Am 19.10.2020 um 17:59 schrieb Mihaela Chirea:

 While going through the token types I also noticed that 
 TOK.arrow seems to be unused. Is there a reason for keeping 
 it? Same for int128 and uns128 but I thought those might be 
 implemented in the future. Is TOK.arrow in a similar situation?
 


 I think the only reason is to give a helpful error message in 
 case someone ports code from C/C++ or writes in that fashion. 
 This is what DMD outputs:

 Error: use . for member lookup, not ->


Actually, at the site where that error is printed the condition 
utilizes 2 distinct tokens [1]: minus and greater than. Therefore 
I think we can get rid of Tok.arrow

[1] https://github.com/dlang/dmd/blob/master/src/dmd/parse.d#L7737
Oct 20 2020
parent =?UTF-8?Q?S=c3=b6nke_Ludwig?= <sludwig outerproduct.org> writes:
Am 21.10.2020 um 04:01 schrieb RazvanN:

 On Monday, 19 October 2020 at 17:37:46 UTC, Sönke Ludwig wrote:

 Am 19.10.2020 um 17:59 schrieb Mihaela Chirea:

 While going through the token types I also noticed that TOK.arrow 
 seems to be unused. Is there a reason for keeping it? Same for int128 
 and uns128 but I thought those might be implemented in the future. Is 
 TOK.arrow in a similar situation?


 I think the only reason is to give a helpful error message in case 
 someone ports code from C/C++ or writes in that fashion. This is what 
 DMD outputs:

 Error: use . for member lookup, not ->


 
 Actually, at the site where that error is printed the condition utilizes 
 2 distinct tokens [1]: minus and greater than. Therefore I think we can 
 get rid of Tok.arrow
 
 [1] https://github.com/dlang/dmd/blob/master/src/dmd/parse.d#L7737


You are right, should also remember to remove TOKarrow then, too.
Oct 20 2020
prev sibling parent Max Haughton <maxhaton gmail.com> writes:
On Sunday, 18 October 2020 at 23:37:04 UTC, Mihaela Chirea wrote:

 Hello!

 Here is an update of the progress made during the last week of 
 Milestone 1 as part of #SAOC2020:
 - I mostly worked on the token location offset tests[1]. This 
 included studying the lexer, parser and tokens in more detail 
 to make sure all the tokens are getting the right offset.
 - Updated the list of locations[2] to show the nodes missing 
 the start location more clearly. It currently contains most 
 nodes in attrib.d and DebugCondition and VersionCondition from 
 cond.d.

 For next week:
 - Add the start location to the nodes that lack this 
 information, starting with the ones from cond.d

 These tasks were mentioned in the previous update[3] and will 
 be done during the second Milestone:
 - Bring all DMD as a library additions hidden behind a version 
 under DMDLIB
 - Add the token size

 The initial milestones can be found here[4]. I will post the 
 updated list as a reply to that thread and use that link in 
 future updates.

 [1]: https://github.com/dlang/dmd/pull/11823
 [2]: 
 https://drive.google.com/file/d/1lb-eP9Pai713tggud6iuzA1cIetbK4zc/view?usp=sharing
 [3]: 
 https://forum.dlang.org/post/vshahvctwdhrmrjssool forum.dlang.org
 [4]: 
 https://forum.dlang.org/post/ztbjfjebxtysmugfzkmz forum.dlang.org


Quick feature suggestion, I've been using DMD to transform code 
to run a safe script inside a game (the details don't matter, 
It's very specific usecase but I may open source it).

It's currently a bit of a pain to do a full semantic analysis 
while also maintaining a compilable buffer i.e. you have to trim 
toPrettyChars() (or whatever it is) to not have the typeinfo 
things (i.e. the mixins from object.d, while also having vanilla 
object.d) for types for when you want to compile your modified 
code. Being able to control this without a flag from a topdown 
view into the compiler (i.e. as a library) could be a big 
stepping stone to fruitful projects.
Oct 19 2020