Replace alloc_struct, free_struct

[stweil]

Both functions simply call malloc, free.

Remove also unneeded null pointer checks and use calloc where possible.

Signed-off-by: Stefan Weil sw@weilnetz.de

[amitdo]

see #11

[amitdo]

Also. it seems that this PR only changes parts of the old engine that Ray wants to drop.

[stweil]

@amitdo, do you think that the PR should not be applied because it changes code which might be removed later?

[amitdo]

@theraysmith commented in #518

Please provide examples of where you get better results with the old engine.
Right now I'm trying to work on getting rid of redundant code, rather than spending time fighting needless changes that generate a lot of work.

Instead of modifying the old engine parts, concentrate on convincing me why it should stay first

To me, the message here from Ray is clear:
<<That's NOT an actual quote of Ray>>
"I don't want you* to do any change to the old engine. Doing so just interfere with my work on the new engine".

That's the way I understand it.

* 'you' - not just you, any developer.

see #11

#11 (review)

[amitdo]

Here is the way I believe things are working in the Google side.

Every commit that we push to this repo, at some time in the future (few weeks/months) will be reviewed by Ray or someone else from Google. If they don't like something they will change it. Then it will be merged with the internal Google branch, which might diverge significantly from the public repo's master branch. I think they do it infrequently, which means that when they try to merge with us they have quite a lot of work to do and it is not so simple as you might think.

@theraysmith, @jbreiden
You are invited to correct me if I'm wrong here.

[theraysmith]

@amitdo, a fair assessment.
Although I haven't made any major changes to the codebase in the last few weeks, yes I would still like to remove the old classifier and take out a lot of code with it.

I'm going to review the replies to my request for "old better than new", and thanks to those that provided them, with a view to making new better than old on those problems. I haven't got to that yet, as I have been totally occupied by fixing many problems I have found with the synthetic training data generation pipeline. Since training takes about 2 weeks, fixes to the source data get priority treatment over other code changes.

Yes, each commit gets reviewed by me and someone else at Google.

[amitdo]

See also:
#752 (comment)

[stweil]

I rebased the code to fix some merge conflicts.

Ray, I noticed your recent changes which introduced TFile in those files. IMO removing memry.{cpp,h} would also be a good step to clean the code. I‌ could do that in two steps to simplify reviews if you agree:

1. switch to using malloc / calloc and free in a first step (like it is done in this PR)
2. replace all those allocations by C++ code (new / new[] and delete / delete[])

[amitdo]

replace all those allocations by C++ code (new / new[] and delete / delete[])

Isn't this remark from Ray also relevant here?
#11 (review)

[stweil]

It isn't, as I promised to use C++ code in step 2 (see above).

[amitdo]

How about replacing the underlying pointer variables with std::array, thus hiding all calls to allocate and delete the memory?

If you know what he meant, consider implementing it.