Alan,
Is there a way to squirrel away a CI build tree that fails for a short
term? Then if someone is interested in a they could inspect the failed
CI build tree? In this case it would make it a lot easier to understand
why it failed rather than having to wait for it to happen again.
Also I'd suggest disabling DEBUG_FEATURES and ERR/WARN/INFO only for
resource constrained configurations. In fact I'd suggest having a CI
build for arch/family that tries to turn everything on to make sure
there are no lurking build failures hiding behind disabled configuration
options.
On 5/6/26 09:49, Alan C. Assis wrote:
Ok, I just tested again enabling:
CONFIG_DEBUG_SYMBOLS=y
CONFIG_DEBUG_SYMBOLS_LEVEL="-g"
And the size didn't change. I think we need to investigate this
"intermittent" issue that happened in the mainline.
But about CONFIG_DEBUG_FEATURES=y and its ERR/WARN/INFO, they will increase
the firmware size, I think those we can remove.
BR,
Alan
On Wed, May 6, 2026 at 10:39 AM raiden00pl <[email protected]> wrote:
-1.
This project is for devs, the main tool used by dev is the debugger and
disabling debug symbols only breaks the workflow for users and makes life
harder.
Debug symbols have no effect on resource usage unless you're loading
ELF directly on target (kernel mode, modules) or something is broken in the
build
system or linker. Debug symbols are DWARF information that get embedded
into
the ELF not binary.
śr., 6 maj 2026 o 15:31 Matteo Golin <[email protected]> napisał(a):
Good idea! In the meantime we can just grep for these options and remove
them.
On Wed, May 6, 2026, 3:27 PM Alan C. Assis <[email protected]> wrote:
Hi Everyone,
Recently we faced an issue that was breaking the CI: the
weact-stm32h750:nsh was exhausting its Flash memory. And it is very
strange
since this STM32H750 chip has 128KB Flash.
My first thought was that some very heavy feature was enabled to nsh
(by
default nsh is supposed to be bare minimum). Then after analysing the
.config I noticed that CONFIG_DEBUG_SYMBOLS=y was enabled.
After disabling it, the Flash usage dropped to 55%. I didn't
investigate
yet what caused the memory use increase in the last few days.
I think we should accept Michael's offer to use his membrowser to see
these
usage increases before the issue triggers the limit. I asked if he
could
release it as a dual license (GPLv3 and Apache) to avoid the concerns
raised on this PR, but didn't receive a response yet.
In the past Greg never allowed DEBUG symbols and features to be enabled
by
default. I think we should keep this same rule because sooner or later
other boards will fail in the same way.
BR,
Alan
