END FUNCTION rewritten as 'end' on ARM64/Termux (Harbour 3.2.0dev r2601171121)

[bertrandsoullier]

Context

• Version: Harbour 3.2.0dev (r2601171121)
• Platform: Termux / Android 15.0 (ARM64, Clang 21.1.8)
• Build options: (Clipper 5.3b) (Clipper 5.x undoc)
• Installation: Compiled from source (make && make install)
• Environment: /data/data/com.termux/files/usr/bin/harbour and hbpp installed correctly

Problem Description
When compiling a simple PRG file containing functions, Harbour produces phantom errors:

clean.prg(22) Error E0010  ENDIF does not match IF
clean.prg(43) Error E0010  ENDIF does not match IF

even though the code is correct and minimal.

Example:

FUNCTION Test()
   IF .T.
      ? "ok"
   ENDIF
RETURN NIL
END FUNCTION

Observed Symptoms

• The generated .ppo file contains lines such as:

command >end<

• END FUNCTION is transformed into end by the internal preprocessor.
• end is then interpreted as ENDIF, which triggers the errors.
• The external preprocessor (hbpp) is fine: when run directly, it does not produce errors.
• The issue persists even after make clean && make install.

Steps to Reproduce

1. Compile Harbour 3.2.0dev (r2601171121) on Termux ARM64/Clang.
2. Create a file clean.prg with a simple function (see example above).
3. Run:

   harbour -p clean.prg
   

4. Observe the ENDIF does not match IF errors.

Expected Result
The code should compile correctly, generating valid .c and .ppo output without errors.

Actual Result
Harbour fails with phantom ENDIF errors.

Hypothesis
The internal preprocessor enabled by default in this revision (Clipper 5.x compatibility options) incorrectly rewrites END FUNCTION.
The bug seems specific to this configuration (Linux/Clang/ARM64).

Possible Solutions

• Disable the internal PP and force Harbour to use hbpp.
• Fix the internal rule that rewrites END FUNCTION.
• Review default build options in this revision.

[alcz]

END FUNCTION is a thing in FoxPro, not Clipper. In Clipper, nor in Harbour there is no rule that rewrites it.
It's the same as anything else like END EXTRAANYTHING

clean.prg(22) Error E0010 ENDIF does not match IF
clean.prg(43) Error E0010 ENDIF does not match IF

You have suspiciously high line counter for a sample that has only six lines.

[woodhead2019]

yes. Clipper and Harbour do not have an 'END FUNCTION' keyword.
Yet AI frequently insert 'END FUNCTION' automatically when generating code.

~/tmp$ cat -n clean.prg
1 FUNCTION Test()
2 IF .T.
3 ? "ok"
4 ENDIF
5 RETURN NIL
6 END FUNCTION

~/tmp$ ~/hb/harbour/bin/linux/gcc/harbour test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
test.prg(6) Error E0010 ENDIF does not match IF
1 error

No code generated.

[bertrandsoullier]

Thank you both for the feedback.
To clarify: the issue is not related to END FUNCTION, FoxPro syntax, or AI‑generated code.

I performed additional tests, and the problem persists even when the source file contains no code at all.

1. Compiling an empty file

echo "" > empty.prg harbour empty.prg

Result:

empty.prg(1) Error E0008 Redefinition of predefined function EMPTY as 'EMPTY'

This happens even though the file contains zero lines.
So the compiler is reading something that is not in the file.

---

2. Compiling a manually written .ppo (bypassing the preprocessor)

FUNCTION Test() RETURN NIL END FUNCTION

Compiled with:

harbour Test.ppo

Result:

Redefinition of procedure or function 'TEST' ENDIF does not match IF

Again, there is no IF, no ENDIF, and only one function.

---

3. External preprocessor (hbpp) injects garbage memory

Running:

hbpp LoadConfig.prg > LoadConfig.ppo

Produces at the top of the file:

ort := 5000

This fragment does not exist in any source file.
It appears to be uninitialized memory being interpreted as code.

---

4. Tested Harbour versions

• master (Jan 2026)
• fd856d4
• 3.2.0 stable
• built with default flags and with -O0

The issue occurs in all cases.

---

Conclusion

This does not appear to be a syntax issue.
It looks like the compiler is reading corrupted or uninitialized memory before processing the first line of the source file.

This would explain:

• phantom ENDIF errors,
• phantom redefinitions,
• incorrect line numbers (22, 43, etc.),
• garbage tokens appearing in .ppo output.

Environment:

• ARM64 (aarch64)
• Android 15 / Termux
• bionic libc
• clang toolchain

Happy to run additional diagnostics or try patches.

[alcz]

I've did Termux build in the past, without noticing such oddities. It was probably GCC though. Does it happen in optimized build too (think you mentioned it already)? If Valgrind is packaged for Termux, you can try to debug memory errors with memcheck tool, typically having Harbour build with those settings

HB_BUILD_DEBUG=yes
HB_BUILD_OPTIM=no

Note, that you probably build Harbour itself in the process on the same platform? Does the problem happen if you build .prg file with hbmk2 test.prg instead of standalone harbour?

[bertrandsoullier]

Thanks for the suggestions.
Yes, the issue happens both in optimized builds and with -O0.
I will rebuild Harbour with:

HBBUILDDEBUG=yes HBBUILDOPTIM=no

and test under Valgrind (if packaged for Termux).
I will also check whether the same error occurs when compiling with hbmk2 test.prg instead of standalone harbour.

Will report back with the results.

[druzus]

Hi, ad1) It's perfect and expected behavior. The same in Clipper and Harbour. HINT: read about -n (or /n) Harbour and Clipper compiler switch. ad2) It's perfect and expected behavior. The same in Clipper and Harbour. HINT: std.ch and #commad END => <x> BTW Passing filename with .ppo extension does not mean that preprocessor is disabled. ad3) hbpp does not output preprocessed file to stdout but creates .ppo file. I have no idea what is inside LoadConfig.prg so I cannot tell you what is the exact reason of such result but for sure _IT_IS_ caused by code inside LoadConfig.prg. best regards, Przemek W dniu 25.01.2026 o 18:10, bertrandsoullier pisze:

…

*bertrandsoullier* left a comment (harbour/core#401) <#401 (comment)> Thank you both for the feedback. To clarify: the issue is not related to END FUNCTION, FoxPro syntax, or AI‑generated code. I performed additional tests, and the problem persists even when the source file contains no code at all. 1. Compiling an empty file |echo "" > empty.prg harbour empty.prg| Result: |empty.prg(1) Error E0008 Redefinition of predefined function EMPTY as 'EMPTY'| This happens even though the file contains zero lines. So the compiler is reading something that is not in the file. ------------------------------------------------------------------------ 2. Compiling a manually written .ppo (bypassing the preprocessor) |FUNCTION Test() RETURN NIL END FUNCTION| Compiled with: |harbour Test.ppo| Result: |Redefinition of procedure or function 'TEST' ENDIF does not match IF| Again, there is no IF, no ENDIF, and only one function. ------------------------------------------------------------------------ 3. External preprocessor (hbpp) injects garbage memory Running: |hbpp LoadConfig.prg > LoadConfig.ppo| Produces at the top of the file: |ort := 5000| This fragment does not exist in any source file. It appears to be uninitialized memory being interpreted as code. ------------------------------------------------------------------------ 4. Tested Harbour versions * master (Jan 2026) * fd856d4 <fd856d4> * 3.2.0 stable * built with default flags and with -O0 The issue occurs in all cases. ------------------------------------------------------------------------ Conclusion This does not appear to be a syntax issue. It looks like the compiler is reading corrupted or uninitialized memory before processing the first line of the source file. This would explain: * phantom ENDIF errors, * phantom redefinitions, * incorrect line numbers (22, 43, etc.), * garbage tokens appearing in .ppo output. Environment: * ARM64 (aarch64) * Android 15 / Termux * bionic libc * clang toolchain Happy to run additional diagnostics or try patches. — Reply to this email directly, view it on GitHub <#401 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA74SHROI72FM6W5LDQ32ST4IT2HPAVCNFSM6AAAAACS2BMS5WVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOOJWHE3DQMZSGY>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[bertrandsoullier]

Update: ASan/UBSan self-build failure
I rebuilt Harbour with:
HB_BUILD_DEBUG=yes
HB_BUILD_OPTIM=no
HB_USER_CFLAGS="-fsanitize=address,undefined -g"
HB_USER_LDFLAGS="-fsanitize=address,undefined"
During the build, the ASan-instrumented harbour crashes while compiling harbinit.prg:
../../../../../bin/linux/gcc/harbour ../../../harbinit.prg -i../../../../../include -n1 -q0 -w3 -es2 -kmo -i-
==6933==WARNING: external symbolizer didn't start up correctly!
make[3]: *** [harbinit.c] Broken pipe
This means that when Harbour is built with AddressSanitizer/UndefinedBehaviorSanitizer, it cannot even compile Harbour itself.
This strongly suggests an internal memory corruption in the compiler.
Combined with previous observations:

• empty.prg (0 bytes) still triggers:
  Error E0008 Redefinition of predefined function EMPTY as 'EMPTY'
• a minimal FUNCTION Test() RETURN NIL END FUNCTION triggers:
  Error E0002 Redefinition of procedure or function 'TEST'
• same behavior with hbmk2
• manually written .ppo files also produce bogus redefinitions
  This happens on:
• ARM64 (aarch64)
• Android 15 / Termux (bionic libc)
• clang toolchain
• Harbour 3.2.0dev (59f2c65 and earlier)
  It looks like the compiler reads corrupted or uninitialized memory before processing the first line of the source file.

[alcz]

I'm aware of some issues pointed out by LLVM sanitizers, they vary in optimized/unoptimized builds, but they fall into undefined behaviour.

../../../../../bin/linux/gcc/harbour ../../../harbinit.prg -i../../../../../include -n1 -q0 -w3 -es2 -kmo -i-
==6933==WARNING: external symbolizer didn't start up correctly!
make[3]: *** [harbinit.c] Broken pipe

^^ this is a problem with the toolchain, not Harbour compiler

This means that when Harbour is built with AddressSanitizer/UndefinedBehaviorSanitizer, it cannot even compile Harbour itself.  
This strongly suggests an internal memory corruption in the compiler.

Untrue.

Example of issues produced by sanitizer in my additional test/build harness:
https://github.com/alcz/hbpksrc/actions/runs/20933381628/job/61506353053#step:9:2131

Combined with previous observations:

- `empty.prg` (0 bytes) still triggers:  
  `Error E0008  Redefinition of predefined function EMPTY as 'EMPTY'`

^^
Choose any other file name than "empty", or use flags mentioned by Przemek.

Clipper (only for reference and curiosity) does not tell about overriding Empty() function:

C:\CLIP53\BIN>clipper empty.prg
CA-Clipper (R) 5.3b (Rev. 338)
Copyright (c) 1985-1996, Computer Associates International, Inc.
Microsoft C Floating Point Support Routines
Copyright (c) Microsoft Corp 1984-1993.  All Rights Reserved.
328K available
Compiling EMPTY.PRG
Code size 32, Symbols 32, Constants 0

C:\CLIP53\BIN>clipper empty.prg /a /n /w /m
CA-Clipper (R) 5.3b (Rev. 338)
Copyright (c) 1985-1996, Computer Associates International, Inc.
Microsoft C Floating Point Support Routines
Copyright (c) Microsoft Corp 1984-1993.  All Rights Reserved.
328K available
Compiling EMPTY.PRG
No code generated

You still have not removed invalid END FUNCTION clause from your sample code.

C:\CLIP53\BIN>clipper clean.prg /a /n /w /m
CA-Clipper (R) 5.3b (Rev. 338)
Copyright (c) 1985-1996, Computer Associates International, Inc.
Microsoft C Floating Point Support Routines
Copyright (c) Microsoft Corp 1984-1993.  All Rights Reserved.
328K available
Compiling CLEAN.PRG
CLEAN.PRG(6)   Error C2027  ENDIF does not match IF
1 error

You're not exactly helpful adding multiple findings under single topic.

I'll retry Termux with clang at some point, to see if I can reproduce it (probably not soon). Maybe you're compiling same .ppo you're writing to, it may be some filesystem oddity under Android machinery, so the file has more lines than expected. If you find the cause, a patch is welcome.

[bertrandsoullier]

Thanks for the feedback.

To avoid mixing unrelated findings, I will keep this ticket focused on
the unexpected redefinition errors with minimal valid source files.

For clarity:

• I acknowledge that EMPTY is a reserved name.
• I also corrected the END FUNCTION issue in my earlier sample.

However, even with a clean file like:

FUNCTION Test()
RETURN NIL
END FUNCTION

I still get:

Error E0002 Redefinition of procedure or function 'TEST'

This happens on Termux/ARM64 (Android 15, clang), even when:

• using hbmk2
• using a static file (not writing to .ppo)
• using different filenames (clean.prg, test1.prg, etc.)

So I will keep this ticket focused on this specific behaviour.

I will open a separate ticket for the ASan/UBSan self-build crash,
since that is a different issue.

[fperillo]

hbmk2 -n test.prg Il Lun 26 Gen 2026, 22:40 bertrandsoullier ***@***.***> ha scritto:

…

*bertrandsoullier* left a comment (harbour/core#401) <#401 (comment)> Thanks for the feedback. To avoid mixing unrelated findings, I will keep this ticket focused on the unexpected redefinition errors with minimal valid source files. For clarity: - I acknowledge that EMPTY is a reserved name. - I also corrected the END FUNCTION issue in my earlier sample. However, even with a clean file like: FUNCTION Test() RETURN NIL END FUNCTION I still get: Error E0002 Redefinition of procedure or function 'TEST' This happens on Termux/ARM64 (Android 15, clang), even when: - using hbmk2 - using a static file (not writing to .ppo) - using different filenames (clean.prg, test1.prg, etc.) So I will keep this ticket focused on this specific behaviour. I will open a separate ticket for the ASan/UBSan self-build crash, since that is a different issue. — Reply to this email directly, view it on GitHub <#401 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA6LSY5ASM2WTLNLLG5BW734I2CWDAVCNFSM6AAAAACS2BMS5WVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQMBRHA4DQNZQGU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[druzus]

W dniu 26.01.2026 o 19:33, Aleksander Czajczyński pisze:

|Combined with previous observations:| |- `empty.prg` (0 bytes) still triggers: `Error E0008 Redefinition of predefined function EMPTY as 'EMPTY'` | ^^ Choose any other file name than "empty", or use flags mentioned by Przemek. Clipper (only for reference and curiosity) does not tell about overriding Empty() function:

Not exactly. Clipper at compile time blocks overriding of all functions which are called by dedicated PCODEs instead od standard function call but the Clipper authors forgot to add this check for functions generated from compiled file names when -n switch is not used and in this example object is created. If you try to compile PRG code with explicit EMPTY() function then you see this error in Clipper. Anyhow in Clipper overridden EMPTY() function cannot be accessed because for all EMPTY() calls Clipper generates PCODEs for direct access internal empty function which is executed instead. best regards, Przemek

[bertrandsoullier]

Thanks for the suggestion.

I tested with:

hbmk2 -n test.prg

using a minimal file:

FUNCTION Test()
RETURN NIL
END FUNCTION

The result is the same:

Error E0002 Redefinition of procedure or function 'TEST'
No code generated.

So the issue also occurs in minimal mode, without .ppo, without .hbp,
without any external preprocessing, and using a static file.

I will continue isolating the behaviour on Termux/ARM64.

[fperillo]

Ok, I did some test on the linux harbour I have, quite old but it doesn't change: a:~/tmp # hbmk2 test.prg Harbour 3.2.0dev (r1601282333) Copyright (c) 1999-2016, http://harbour-project.org/ Compiling 'test.prg'... test.prg(4) Error E0030 Syntax error "syntax error at 'END'" 1 error No code generated. hbmk2: Error: Running Harbour compiler (embedded). 1 (/usr/local/src/harbour/bin/linux/gcc/harbour) *-n2* test.prg -o/tmp/hbmk_UW5g85.dir/ -i/usr/local/src/harbour/include *Please note that hbmk2 adds -n2 parameter automatically, in my setup, both in linux and in windows* a:~/tmp # harbour test.prg Harbour 3.2.0dev (r1601282333) Copyright (c) 1999-2016, http://harbour-project.org/ Compiling 'test.prg'... test.prg(2) Error E0002 Redefinition of procedure or function 'TEST' test.prg(4) Error E0030 Syntax error "syntax error at 'END'" 2 errors No code generated. *Calling harbour compiler directly, no -n parameter, redefinition error is raised AS IT SHOULD BE* a:~/tmp # harbour -n test.prg Harbour 3.2.0dev (r1601282333) Copyright (c) 1999-2016, http://harbour-project.org/ Compiling 'test.prg'... test.prg(4) Error E0030 Syntax error "syntax error at 'END'" 1 error No code generated. Adding the -n parameter harbour compiler doesn't auto-generate a function/procedure based on the filename and correctly compiles the function in the source code. The error at END is CORRECT So please tell use what hbmk2 reports, like mine reports: hbmk2: Error: Running Harbour compiler (embedded). 1 (/usr/local/src/harbour/bin/linux/gcc/harbour) *-n2* test.prg -o/tmp/hbmk_UW5g85.dir/ -i/usr/local/src/harbour/include And please remove END FUNCTION that is not a valid clipper statement and there is nothing open to close...

…

On Tue, Jan 27, 2026 at 11:30 AM bertrandsoullier ***@***.***> wrote: *bertrandsoullier* left a comment (harbour/core#401) <#401 (comment)> Thanks for the suggestion. I tested with: hbmk2 -n test.prg using a minimal file: FUNCTION Test() RETURN NIL END FUNCTION The result is the same: Error E0002 Redefinition of procedure or function 'TEST' No code generated. So the issue also occurs in minimal mode, without .ppo, without .hbp, without any external preprocessing, and using a static file. I will continue isolating the behaviour on Termux/ARM64. — Reply to this email directly, view it on GitHub <#401 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA6LSY2RIHCG255SS5D6I434I442RAVCNFSM6AAAAACS2BMS5WVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQMBUGM3DQNBQGE> . You are receiving this because you commented.Message ID: ***@***.***>

[bertrandsoullier]

Thanks for the detailed explanation.

Here is what hbmk2 reports on my Termux/ARM64 setup:

hbmk2 test.prg
Harbour 3.2.0dev (...)
Compiling 'test.prg'...
test.prg(2) Error E0002 Redefinition of procedure or function 'TEST'
No code generated.
hbmk2: Error: Running Harbour compiler (embedded). 1
(/data/data/com.termux/files/usr/bin/harbour) -n2 test.prg
-o/tmp/hbmk_xxxxx.dir/ -i/data/data/com.termux/files/usr/include

So hbmk2 also injects the -n2 parameter here, exactly as in your setup.

For clarity, I also tested with a minimal file without END FUNCTION:

FUNCTION Test()
RETURN NIL

harbour test.prg → E0002 (redefinition)
harbour -n test.prg → E0002 (redefinition)
hbmk2 test.prg → E0002 (redefinition)

In other words, on this platform the redefinition error is raised
even in minimal mode, with -n or -n2, and with a syntactically valid
single-function source file.

This is why I initially suspected a platform‑specific issue on
Termux/ARM64 rather than a syntax problem in the test file.

[alcz]

But it is not general ARM64/Android platform issue, I've rebuilt Harbour with the same toolchain. Modern Android versions/devices may employ syscall filtering and could have quirks in the filesystem, but why should you encounter it in such basics?

Considering an older Android device, but with updated Termux...

Before updating Termux packages:

~/.../linux/clang $ cat test.prg
FUNCTION Test()
RETURN NIL

~/.../linux/clang $ ./harbour test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
test.prg(1) Error E0002  Redefinition of procedure or function 'TEST'
1 error

No code generated.

~/.../linux/clang $ ./harbour -n test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
Lines 2, Functions/Procedures 1
Generating C source output to 'test.c'... Done.
~/.../linux/clang $ ./hbmk2 test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
Lines 2, Functions/Procedures 1
Generating C source output to '/data/data/com.termux/files/usr/tmp/hbmk_0oc9e9.dir/test.c'... Done.
~/.../linux/clang $ ./harbour -build
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/

Harbour Build Info
---------------------------
Version: Harbour 3.2.0dev (r2601171121)
Compiler: LLVM/Clang C 19.1.6 ARM64
Platform: Linux 4.19.136-perf+ aarch64
PCode version: 0.3
ChangeLog last entry: 2026-01-17 12:21 UTC+0100 Przemyslaw Czerpak (druzus/at/poczta.onet.pl)
ChangeLog ID: 6362ce0359a9f8830e8e5974994ca11ac6043c9c
Built on: Jan 27 2026 15:52:28
Build options: (Clipper 5.3b) (Clipper 5.x undoc)
---------------------------

After upgrading Termux packages, make clean && make:

~/.../linux/clang $ ./harbour test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
test.prg(1) Error E0002  Redefinition of procedure or function 'TEST'
1 error

No code generated.
~/.../linux/clang $ ./harbour -n test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
Lines 2, Functions/Procedures 1
Generating C source output to 'test.c'... Done.
~/.../linux/clang $ ./hbmk2 test.prg
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/
Compiling 'test.prg'...
Lines 2, Functions/Procedures 1
Generating C source output to '/data/data/com.termux/files/usr/tmp/hbmk_er4ew4.dir/test.c'... Done.

~/.../linux/clang $ ./harbour -build
Harbour 3.2.0dev (r2601171121)
Copyright (c) 1999-2025, https://harbour.github.io/

Harbour Build Info
---------------------------
Version: Harbour 3.2.0dev (r2601171121)
Compiler: LLVM/Clang C 21.1.8 ARM64
Platform: Linux 4.19.136-perf+ aarch64
PCode version: 0.3
ChangeLog last entry: 2026-01-17 12:21 UTC+0100 Przemyslaw Czerpak (druzus/at/poczta.onet.pl)
ChangeLog ID: 6362ce0359a9f8830e8e5974994ca11ac6043c9c
Built on: Jan 27 2026 16:45:15
Build options: (Clipper 5.3b) (Clipper 5.x undoc)
---------------------------