[llvm-dev] [RFC][FileCheck] New option to negate check patterns

Robinson, Paul via llvm-dev llvm-dev at lists.llvm.org
Fri Jan 31 06:31:58 PST 2020

Seems to me there was a proposal (possibly years ago now) to allow defining named patterns, from someone who had (IIRC) implemented such a feature in a downstream project.  I don’t remember the details of their use-case, but apparently by itself it wasn’t compelling enough to get the encouragement to proceed.

FileCheck directives are already easy to get wrong, and adding a command-line option to change their interpretation just seems like asking for trouble. But adding a way to define a pattern that is independent of the input text seems like it could be useful.

I’d suggest that initially at least, the define-a-pattern directive would take only “immediate” text, no embedded regexes. That is, you could do
    DEFPAT[MYPATTERN]: Define a pattern here
but you couldn’t do
    DEFPAT[MYPATTERN]: Define {{some|any}} pattern here
although it might be reasonable to allow
    DEFPAT[PATTERN1]: some
    DEFPAT[PATTERN2]: Define [[PATTERN1]] pattern here
as the [[]] substitution can be done when the directive is read.

My $.02,

From: George Rimar <grimar at accesssoftek.com>
Sent: Friday, January 31, 2020 5:52 AM
To: Thomas Preud'homme <thomasp at graphcore.ai>; llvm-dev <llvm-dev at lists.llvm.org>; Robinson, Paul <paul.robinson at sony.com>; jh7370.2008 at my.bristol.ac.uk
Subject: Re: [RFC][FileCheck] New option to negate check patterns

​Hi all,

> I feel it might be confusing to have a CHECK becomes effectively a CHECK-NOT,

> especially if the RUN line is far from the CHECK line (which is often the case when

> a single RUN line drives several groups of CHECK directives (e.g. code generation

> tested for several functions for a specific feature, like PIC). You also loose control

> on where the NOT should be: it would have to be at the same location as the

> CHECK even though for the NOT case you might want to check it somewhere else.

I think I agree with Thomas.

+ the relationship with "CHECK-NOT/CHECK-NEXT/CHECK-SAME" might make things overcomplicated probably.

> How about having a concept of regex variables where you give a name
> to a given directive's pattern which you could reuse as another pattern. Something like (syntax TBD):
> CHECK<NAME>: mov [[REG:r[0-9]+]], #42

I.e. without adding a new optinons for FileCheck, something like the following?
# RUN: llvm-sometool --print-string | FileCheck %s --check-prefix=CHECK1
# RUN: llvm-sometool --no-print-string | FileCheck %s --check-prefix=CHECK2

CHECK1<NAME>: mov [[REG:r[0-9]+]], #42

It might work probably.  We already have the ability to name parts of
the output checked:

// CHECK:      Dynamic Relocations {
// CHECK-NEXT:   {{.*}} R_AARCH64_RELATIVE - [[BAR_ADDR:.*]]

// CHECK:      Symbols [
// CHECK-NEXT:     Value: [[BAR_ADDR]]

So adding a way for naming the whole line does not look
an unreasonable/inconsistent extention to me I think.

Best regards,​
George | Developer | Access Softek, Inc​​​​​​​

From: James Henderson <jh7370.2008 at my.bristol.ac.uk>
Sent: 31 January 2020 09:14
To: llvm-dev <llvm-dev at lists.llvm.org>; Thomas Preud'homme <thomasp at graphcore.ai>; Paul Robinson <paul.robinson at sony.com>; George Rimar <grimar at accesssoftek.com>
Subject: [RFC][FileCheck] New option to negate check patterns

     [This message was sent from somebody outside of your organisation]

Hi all,

There have been a few cases recently where I've noticed two test cases in the same lit test that do the same thing except invert the CHECK, to show that something is NOT present. I'm talking about something like the following:

# RUN: llvm-sometool --print-string | FileCheck %s --check-prefix=STRING
# RUN: llvm-sometool --no-print-string | FileCheck %s --check-prefix=NO-STRING
# STRING: This is the string
# NO-STRING-NOT: This is the string

In such cases, as can be seen, the CHECK line effectively has to be duplicated (either in an explicit check like in the above example, or via --implicit-check-not). Duplication is generally bad, especially in this sort of case, as it only takes a typo in the NOT pattern, or a careless developer/reviewer pair changing the output to cause the NOT pattern to no longer be useful.

I'd like to propose a new FileCheck option (e.g. --check-not-prefix/--check-not-prefixes) which allows implicitly converting a check prefix to a -NOT version of the same prefix. That would allow writing the above example as:

# RUN: llvm-sometool --print-string | FileCheck %s --check-prefix=STRING
# RUN: llvm-sometool --no-print-string | FileCheck %s --check-not-prefix=STRING
# STRING: This is the string

If there was a typo or somebody changed the string output, this mechanism would ensure there is no chance of the pattern rotting.

Caveat: I don't know what would be the appropriate way of handling non-trivial checks, i.e. existing CHECK-NOT/CHECK-NEXT/CHECK-SAME etc. I'd appreciate any ideas on this.



** We have updated our privacy policy, which contains important information about how we collect and process your personal data. To read the policy, please click here<http://www.graphcore.ai/privacy> **

This email and its attachments are intended solely for the addressed recipients and may contain confidential or legally privileged information.
If you are not the intended recipient you must not copy, distribute or disseminate this email in any way; to do so may be unlawful.

Any personal data/special category personal data herein are processed in accordance with UK data protection legislation.
All associated feasible security measures are in place. Further details are available from the Privacy Notice on the website and/or from the Company.

Graphcore Limited (registered in England and Wales with registration number 10185006) is registered at 107 Cheapside, London, UK, EC2V 6DN.
This message was scanned for viruses upon transmission. However Graphcore accepts no liability for any such transmission.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200131/eb058e50/attachment-0001.html>

More information about the llvm-dev mailing list