Add assert statement to Fortran #70

certik · 2019-11-04T23:25:36Z

One very common use case for a preprocessor is a custom ASSERT macro, that is empty in Release mode, but in Debug mode it checks the condition and prints out the filename and line number. As discussed in #65, Fortran is moving away from a preprocessor, and so by adding an assert statement into the language would eliminate one more use case for a preprocessor.

From a user perspective, the syntax can be something like:

integer :: i
i = 5
assert(i == 5)

In Release mode, the assert would do nothing, in Debug mode the assert would evaluate the condition (e.g., i == 5) and if it fails, it would print a nice stacktrace. This would be a much better behavior than the current practice, where typically the ASSERT macro only prints a filename+line, but not a full stacktrace, so it can be hard to figure out why it happened.

The text was updated successfully, but these errors were encountered:

sblionel · 2019-11-04T23:49:33Z

Assert has been proposed before (https://j3-fortran.org/doc/year/04/04-376.txt) This was before my time on the committee, but I gather from what evidence I can find that there was not sufficient interest to move it forward. I have seen it implemented as a vendor extension.

certik · 2019-11-05T00:12:41Z

@sblionel thanks! I would like for each idea at this GitHub repository to eventually have a technical response from the committee why it was rejected. And if there is no technical reason, just a lack of sufficient interest at some point in the past, then I think we can revisit such an idea. For example in this particular case in light of #65 we can argue that rather than standardizing a preprocessor, let's improve Fortran so that the preprocessor is not needed (as much).

klausler · 2019-11-05T00:37:10Z

ASSERT is also a potential optimization aid:

  ASSERT (K >= 0)
  DO J = 1, N
    A(J) = A(J + K)  ! known to not cause a data dependence cycle
  END DO

aradi · 2019-11-12T16:45:38Z

Probably, ASSERT should not be used for optimization. As evaluation of an assert could be time consuming (imagine ASSERT(all(big_array > 0)), I'd wish that it is not considered when the code is compiled for production. Therefore, it could even be a pre-processor feature instead of a compiler feature (see also #65).

certik · 2019-11-12T16:47:47Z

@aradi the proposal here, as I understand it, is that ASSERT(all(big_array > 0)) will do nothing in Release mode. But I do not understand your argument why the compiler couldn't still use it for optimization.

sblionel · 2019-11-12T16:55:51Z

DEC implemented ASSERT as follows: It is a declaration from the programmer to the optimizer that the given condition is met. The optimizer can then use that information. A separate option tells the compiler to generate run-time code to check that assertion, so you can choose to do that or not.

Intel Fortran currently supports an ASSUME directive that provides information, but there is not currently any run-time code associated with that.

aradi · 2019-11-12T17:01:06Z

@certik What happens, if in production mode (for whatever buggy reasons) the criteria is not met. The run-time code does not check for it, but it was taken into account during the optimization. Then the optimized code may produce results which are hard to understand/debug. But on the other hand, probably you can say, it is the programmers fault...

certik · 2019-11-12T17:24:31Z

@certik What happens, if in production mode (for whatever buggy reasons) the criteria is not met. The run-time code does not check for it, but it was taken into account during the optimization. Then the optimized code may produce results which are hard to understand/debug. But on the other hand, probably you can say, it is the programmers fault...

Yes, it would give wrong results in Release mode, but not in Debug mode. In exactly the same way as you get wrong results if you go out of bounds (in Release mode). But in Debug mode, out of bounds, as well as the assert conditions are all checked.

aradi · 2019-11-12T20:19:36Z

@certik OK, that's a good point. It makes sense, I agree, and would be consistent with current compiler behavior as well.

rweed · 2020-01-03T14:49:09Z

If we can't have an ASSERT statement, I would be in favor of a standard assert subroutine. You can then use the pre-processor to create a macro ASSERT (or ERROR_ABORT) command. Here is my version that I "borrowed" from the FTL project

! Macros for assertions

#ifdef GFORTRAN
#define ASSERT(x) call assert(x,"x",FILE,LINE)
#define ERROR_ABORT(x,y) call assert(x,"x",FILE,LINE,y)
#else
#define ASSERT(x) call assert(x,#x,FILE,LINE)
#define ERROR_ABORT(x,y) call assert(x,#x,FILE,LINE,y)
#endif

ERROR_ABORT just adds a logical to signal if you want to do a hard aborut (STOP,
ERROR_STOP) or continue after printing an error message

certik · 2020-01-03T14:56:33Z

@rweed I think this would be a perfect addition to stdlib. Can you please open up an issue there? We already have a minimal assert subroutine there:

https://github.com/fortran-lang/stdlib/blob/924ee5459fab5c9fa4deb30e75c12c00b2864598/src/stdlib_experimental_error.f90#L8

So let's extend it, and then let's discuss the macros also, per your comment.

wclodius2 · 2020-08-01T21:09:05Z

C++ 2020 and Ada (2012?) defined three statements intended to make "Design by Contract" (DbC) an intrinsic part of the language. In C++ these three statements define the attributes: requires, asserts, and ensures, where requires and ensures are in effect assert statements that are executed upon subprogram entry and exit respectively. requires, in effect, defines the pre-conditions the caller must meet for correct operation of the subprogram, and ensures the post-conditions the caller can expect to be satisfied by the subprogram if it has operated correctly.

I have written a draft proposal for the incorporation of DbC in Fortran 202y. It, of course, includes an ASSERT statement as part of the proposal. If there is interest I can upload the proposal to my branch of the proposals directory. Once there, given the problems with my last PR, where it got added on to an existing PR, I don't know if I should do a PR or have someone else do the PR for me.

certik · 2020-08-01T21:11:30Z

@wclodius2 Yes, please upload it here. Try to create a new branch and put it there. Once you learn the process, it becomes simple. I can help.

wclodius2 · 2020-08-01T21:39:12Z

GitHub doesn’t let me do an additional fork on https://github.com/wclodius2/fortran_proposals <https://github.com/wclodius2/fortran_proposals>, but does let me do a fork on the original https://github.com/j3-fortran/fortran_proposals <https://github.com/j3-fortran/fortran_proposals>. Should I fork the original again?

…

On Aug 1, 2020, at 3:11 PM, Ondřej Čertík ***@***.***> wrote: @wclodius2 <https://github.com/wclodius2> Yes, please upload it here. Try to create a new branch and put it there. Once you learn the process, it becomes simple. I can help. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#70 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APTQDORJEOLRGNHAYVX7WHDR6SAI7ANCNFSM4JI3NFHQ>.

wclodius2 · 2020-08-02T16:48:16Z

FWIW in not letting me fork on my repository it tells me "Cannot fork because you own this repository and are not a member of any organization."

certik · 2020-08-04T14:53:32Z

@wclodius2 here is how you do it. You already have a fork at GitHub (https://github.com/wclodius2/fortran_proposals) so that is all you need. Let's start from scratch on your computer, that way things will work exactly as I write them here. You will see that once you learn this workflow, you can apply it to any other project at GitHub.

Go to some place on your computer where you want to have the fortran_proposals directory. Then initialize it as follows:

git clone https://github.com/j3-fortran/fortran_proposals
cd fortran_proposals
git remote add william [email protected]:wclodius2/fortran_proposals.git

Then create a branch assert (you can name it any other way) and put your document in:

git checkout -b assert
# copy your document to proposals/assert, let's call it `assert.pdf` (I don't know what you call it)
git add proposals/assert/assert.pdf
git commit # an editor will open, write a commit message
git push william

This will push the assert branch into your fork, and it will also give you a link that will allow you to submit the Pull Request (PR).

Let me know if this works, and if something does not work, I'll help.

wclodius2 · 2020-08-04T16:21:38Z

Before I do anything irreversible what is the meaning of <name> in git remote add <name> <url> where you have put william for <name>, i.e., is it arbitrary or should it be my user name on my computer or something else?

…

On Aug 4, 2020, at 8:53 AM, Ondřej Čertík ***@***.***> wrote: @wclodius2 <https://github.com/wclodius2> here is how you do it. You already have a fork at GitHub (https://github.com/wclodius2/fortran_proposals <https://github.com/wclodius2/fortran_proposals>) so that is all you need. Let's start from scratch on your computer, that way things will work exactly as I write them here. You will see that once you learn this workflow, you can apply it to any other project at GitHub. Go to some place on your computer where you want to have the fortran_proposals directory. Then initialize it as follows: git clone https://github.com/j3-fortran/fortran_proposals cd fortran_proposals git remote add william ***@***.***:wclodius2/fortran_proposals.git Then create a branch assert (you can name it any other way) and put your document in: git checkout -b assert # copy your document to proposals/assert, let's call it `assert.pdf` (I don't know what you call it) git add proposals/assert/assert.pdf git commit # an editor will open, write a commit message git push william This will push the assert branch into your fork, and it will also give you a link that will allow you to submit the Pull Request (PR). Let me know if this works, and if something does not work, I'll help. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#70 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APTQDOTH72WGDCOILTM5PY3R7AOH3ANCNFSM4JI3NFHQ>.

certik · 2020-08-04T16:25:21Z

The meaning of <name> is an arbitrary identifier of your choosing that you then reference in git push. When you do git remote -v, it will show what is setup. And you can always rename <name> using git remote rename william something_else. Here is how I have it setup for myself:

ondrej@pn1707483:~/repos/fortran_proposals(master)$ git remote -v
ondrej	[email protected]:certik/fortran_proposals (fetch)
ondrej	[email protected]:certik/fortran_proposals (push)
origin	https://github.com/lfortran/fortran_proposals (fetch)
origin	https://github.com/lfortran/fortran_proposals (push)

Which is analogous to what I suggested for you. If I add william to mine, see what happens:

ondrej@pn1707483:~/repos/fortran_proposals(master)$ git remote add william [email protected]:wclodius2/fortran_proposals.git
ondrej@pn1707483:~/repos/fortran_proposals(master)$ git remote -v
ondrej	[email protected]:certik/fortran_proposals (fetch)
ondrej	[email protected]:certik/fortran_proposals (push)
origin	https://github.com/lfortran/fortran_proposals (fetch)
origin	https://github.com/lfortran/fortran_proposals (push)
william	[email protected]:wclodius2/fortran_proposals.git (fetch)
william	[email protected]:wclodius2/fortran_proposals.git (push)

Now I can pull branches from you using git fetch william. So that's the idea.

wclodius2 · 2020-08-04T18:29:56Z

On the git push william I get The authenticity of host 'github.com (140.82.114.3)' can't be established. RSA key fingerprint is SHA256:nThbg6kXUpJWGl7E1IGOCspRomTxdCARLviKw6E5SY8. Are you sure you want to continue connecting (yes/no/[fingerprint])? yes Warning: Permanently added 'github.com,140.82.114.3' (RSA) to the list of known hosts. [email protected]: Permission denied (publickey). fatal: Could not read from remote repository. Please make sure you have the correct access rights and the repository exists.

…

On Aug 4, 2020, at 8:53 AM, Ondřej Čertík ***@***.***> wrote: @wclodius2 <https://github.com/wclodius2> here is how you do it. You already have a fork at GitHub (https://github.com/wclodius2/fortran_proposals <https://github.com/wclodius2/fortran_proposals>) so that is all you need. Let's start from scratch on your computer, that way things will work exactly as I write them here. You will see that once you learn this workflow, you can apply it to any other project at GitHub. Go to some place on your computer where you want to have the fortran_proposals directory. Then initialize it as follows: git clone https://github.com/j3-fortran/fortran_proposals cd fortran_proposals git remote add william ***@***.***:wclodius2/fortran_proposals.git Then create a branch assert (you can name it any other way) and put your document in: git checkout -b assert # copy your document to proposals/assert, let's call it `assert.pdf` (I don't know what you call it) git add proposals/assert/assert.pdf git commit # an editor will open, write a commit message git push william This will push the assert branch into your fork, and it will also give you a link that will allow you to submit the Pull Request (PR). Let me know if this works, and if something does not work, I'll help. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#70 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APTQDOTH72WGDCOILTM5PY3R7AOH3ANCNFSM4JI3NFHQ>.

certik · 2020-08-04T19:03:11Z

That means your ssh key is not at GitHub (I assume it already was there). All you need to do is to create an ssh key:

ssh-keygen

and copy ~/.ssh/id_rsa.pub (assuming you are on linux) to your GitHub profile: https://github.com/settings/keys

The new paper, design_by_contract.txt, contains a proposal to allow Fortran programmers to use "Design by Contract", proposing three new statements for the Fortran language: REQUIRES, ASSERT, and ENSURES. As such, among other things, it addresses issue j3-fortran#70 'Add assert statement to Fortran'. [ticket: j3-fortran#70]

In reviewing the paper prior to submitting a pull request I realized that I had misspelled .EQV. as .EQUIV. and that I had forgotten to mention ASSERT at one point where I discussed REQUIRES and ENSURES. [ticket: j3-fortran#70]

Changed ASSERTS, ENSURES, and REQUIRES to ASSERT, ENSURE, and REQUIRE, respectively [ticket: j3-fortran#70]

rouson · 2023-04-07T21:15:17Z

@wclodius2 I would like to see your proposal included in the Fortran 202Y work list if it's not too late. It look like the above comment thread trailed off without resolution of how to contribute the proposal. If I help, my preference would be to do so interactively via Zoom or something similar because I'm more likely to be able to help if there's dedicated synchronous time than if we bounce back and forth asynchronously. @certik would you like to join such a call if it happens?

wclodius2 · 2023-04-08T04:04:32Z

Feel free to use it. I assume that there is still a copu on the site. I have been into other things the past year and have not kept track of the progress on the project. I had assumed that there was not much interest in that work as the effort seemed to have gone in a rather different direction. Sent from my iPhoneOn Apr 7, 2023, at 3:15 PM, Damian Rouson ***@***.***> wrote: @wclodius2 I would like to see your proposal included in the Fortran 202Y work list if it's not too late. It look like the above comment thread trailed off without resolution of how to contribute the proposal. If I help, my preference would be to do so interactively via Zoom or something similar because I'm more likely to be able to help if there's dedicated synchronous time than if we bounce back and forth asynchronously. @certik would you like to join such a call if it happens? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: ***@***.***>

shahmoradi · 2023-04-08T06:18:07Z

Our research library currently contains over 1300 FPP ASSERT macros that can be (de)activated under any compilation mode. This has proven vital to the health of the codebase and catching tens of intricate bugs that could have remained undetected.
While adding an intrinsic assert statement to the language will resolve most of our use cases, this appears more of a piecemeal approach that does not resolve the more significant issue: the need for macros or similar concepts in Fortran. There is not a single day that I do not wish Fortran had macros beyond what can be currently done with FPP.

tkoenig1 · 2023-04-08T07:35:18Z

@shahmoradi

hile adding an intrinsic assert statement to the language will resolve most of our use cases, this appears more of a piecemeal approach that does not resolve the more significant issue: the need for macros or similar concepts in Fortran. There is not a single day that I do not wish Fortran had macros beyond what can be currently done with FPP.

This was one of the reasons why I submitted issue #293 , this was motivated by people using macro preprocessors. Generics know the language, and can be much more expessive than (rather simple-minded) macros.

Could you maybe list your use cases there, so we could try to formulate them there?

shahmoradi · 2023-04-09T03:10:37Z

@tkoenig1 #293 is close to what I dreamed of having in Fortran someday, hopefully very soon. I encounter many instances almost daily where macros appear as the only solution, to my knowledge (with the alternative being code duplication). I am not talking about generic types here, merely code patterns. For example, many logical branches of BLAS/LAPACK routines that I have inspected differ only in some array indices order, e.g., (i, j) vs. (j, i). With macros or similar functionality, the size of those routines could be reduced multifold, making them more readable and maintainable with fewer opportunities for introducing bugs. Add generic programming to the routines to get another multifold size reduction for BLAS/LAPACK libraries and similar.

klausler · 2023-04-28T17:56:02Z

A compile-time static assert macro:

$ cat assert.F90
#define STATIC_ASSERT(x) block; real(merge(kind(1.),-1,(x))), parameter :: __fail = 1.; end block
STATIC_ASSERT(1 > 2)
end
$ f18 assert.F90
assert.F90:2:1: error: REAL(KIND=-1) is not a supported type
  STATIC_ASSERT(1 > 2)
  ^^^^^^^^^^^^^^^^^^^^
assert.F90:1:26: in a macro defined here
  #define STATIC_ASSERT(x) block; real(merge(kind(1.),-1,(x))), parameter :: __fail = 1.; end block
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
that expanded to:
  block; real(merge(kind(1.),-1,(1 > 2))), parameter :: __fail = 1.; end block
         ^

certik · 2023-04-28T18:24:40Z

@klausler clever! Here is how to do with with gfortran:

$ cat a.f90 
#define STATIC_ASSERT(x) block; real(merge(kind(1.),-1,(x))), parameter :: fail = 1.; end block
!STATIC_ASSERT(1 > 2)
STATIC_ASSERT(sin(0.5) > 0.5)
end
$ gfortran -cpp a.f90
a.f90:3:48:

    3 | STATIC_ASSERT(sin(0.5) > 0.5)
      |                                                1
Error: Kind -1 not supported for type REAL at (1)

ivan-pi · 2023-04-28T19:00:55Z

You can even add a message:

define STATIC_ASSERT(x,msg) block; character(kind=merge(kind('a'),-1,(x))), parameter :: fail = msg; end block

$ gfortran -cpp -ffree-line-length-none assert.f90 
assert.f90:2:69:

    2 | STATIC_ASSERT(storage_size(1.d0) >= 80,"double precision must be at least 80-bit")
      |                                                                     1
Error: Kind -1 is not supported for CHARACTER at (1)
$ gfortran -fdefault-real-16 -cpp -ffree-line-length-none assert.f90 
$

certik mentioned this issue Nov 4, 2019

The preprocessor should be standardized #65

Open

certik mentioned this issue Nov 11, 2019

Optional applied to groups of arguments #77

Open

ivan-pi mentioned this issue Nov 15, 2019

Optional dimensioning in assumed-shape arrays #89

Open

rweed mentioned this issue Jan 3, 2020

Proposals for February 24 - 28, 2020 J3 committee meeting #122

Open

certik mentioned this issue Jan 21, 2020

Pass optional message to assert fortran-lang/stdlib#116

Closed

certik mentioned this issue Mar 23, 2020

Proposal for test_condition fortran-lang/stdlib#121

Closed

wclodius2 added a commit to wclodius2/fortran_proposals that referenced this issue Aug 5, 2020

Changed to active voice

9ae1081

Changed ASSERTS, ENSURES, and REQUIRES to ASSERT, ENSURE, and REQUIRE, respectively [ticket: j3-fortran#70]

MarDiehl mentioned this issue Jun 17, 2021

Formulate long term vision for Fortran in terms of features #59

Open

certik added the Clause 10 Standard Clause 10: Expressions and assignment label Apr 23, 2022

certik mentioned this issue Apr 26, 2024

Assert statement lfortran/lfortran#3941

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add assert statement to Fortran #70

Add assert statement to Fortran #70

certik commented Nov 4, 2019 •

edited

Loading

sblionel commented Nov 4, 2019

certik commented Nov 5, 2019 •

edited

Loading

klausler commented Nov 5, 2019

aradi commented Nov 12, 2019

certik commented Nov 12, 2019

sblionel commented Nov 12, 2019 •

edited

Loading

aradi commented Nov 12, 2019

certik commented Nov 12, 2019

aradi commented Nov 12, 2019

rweed commented Jan 3, 2020

certik commented Jan 3, 2020

wclodius2 commented Aug 1, 2020

certik commented Aug 1, 2020

wclodius2 commented Aug 1, 2020 via email

wclodius2 commented Aug 2, 2020

certik commented Aug 4, 2020

wclodius2 commented Aug 4, 2020 via email

certik commented Aug 4, 2020 •

edited

Loading

wclodius2 commented Aug 4, 2020 via email

certik commented Aug 4, 2020

rouson commented Apr 7, 2023

wclodius2 commented Apr 8, 2023 via email

shahmoradi commented Apr 8, 2023

tkoenig1 commented Apr 8, 2023

shahmoradi commented Apr 9, 2023

klausler commented Apr 28, 2023

certik commented Apr 28, 2023

ivan-pi commented Apr 28, 2023 •

edited

Loading

Add assert statement to Fortran #70

Add assert statement to Fortran #70

Comments

certik commented Nov 4, 2019 • edited Loading

sblionel commented Nov 4, 2019

certik commented Nov 5, 2019 • edited Loading

klausler commented Nov 5, 2019

aradi commented Nov 12, 2019

certik commented Nov 12, 2019

sblionel commented Nov 12, 2019 • edited Loading

aradi commented Nov 12, 2019

certik commented Nov 12, 2019

aradi commented Nov 12, 2019

rweed commented Jan 3, 2020

certik commented Jan 3, 2020

wclodius2 commented Aug 1, 2020

certik commented Aug 1, 2020

wclodius2 commented Aug 1, 2020 via email

wclodius2 commented Aug 2, 2020

certik commented Aug 4, 2020

wclodius2 commented Aug 4, 2020 via email

certik commented Aug 4, 2020 • edited Loading

wclodius2 commented Aug 4, 2020 via email

certik commented Aug 4, 2020

rouson commented Apr 7, 2023

wclodius2 commented Apr 8, 2023 via email

shahmoradi commented Apr 8, 2023

tkoenig1 commented Apr 8, 2023

shahmoradi commented Apr 9, 2023

klausler commented Apr 28, 2023

certik commented Apr 28, 2023

ivan-pi commented Apr 28, 2023 • edited Loading

certik commented Nov 4, 2019 •

edited

Loading

certik commented Nov 5, 2019 •

edited

Loading

sblionel commented Nov 12, 2019 •

edited

Loading

certik commented Aug 4, 2020 •

edited

Loading

ivan-pi commented Apr 28, 2023 •

edited

Loading