jedrzejboczar / elf-size-analyze Goto Github PK

I have rather long typenames due to template types and namespaces, and the default printing method cuts the symbol names, to the point I can't distinguish them.

======================================== ROM =========================================
Symbol                                                              Size        %     
======================================================================================
/my/project/path/src/util                      2.3 KiB   100.00
  convert                                                             1.7 KiB    73.03
    CompactParse.cc                                                   1.1 KiB    45.40
      pj_impl_1_0_0::tryParseNumber(char const*)                      1.1 KiB    45.40
    UTF16Parse.cc                                                   664.0 B      27.63
      writeCodeunit(char*, int)                                     303.0 B      12.61
      pj_impl_1_0_0::tryConvertInplace(char*, bool)                 242.0 B      10.07
      parseHex4Unchecked(char const*)                               119.0 B       4.95
  arena/DynArenaImpl.cc                                             648.0 B      26.97
    pj_impl_1_0_0::DynArenaImpl::setAvailable(unsigned short, ...   332.0 B      13.82
    pj_impl_1_0_0::DynArenaImpl::resizeAlloc(unsigned long, un...   110.0 B       4.58
    pj_impl_1_0_0::DynArenaImpl::DynArenaImpl(pj_impl_1_0_0::D...   103.0 B       4.29
    pj_impl_1_0_0::DynArenaImpl::DynArenaImpl(pj_impl_1_0_0::D...   103.0 B       4.29
?                                                                     0.0 B       0.00
======================================================================================
Symbols total                                                         2.3 KiB         
======================================================================================

(Note the last 2 results)

Could you please add an option to use the longest symbol name to calculate the table width and as such print all full symbol names?

Hi @jedrzejboczar
I forked this awesome script and refactored it, added json support and created a project out of it. Now, after some more beautifying and adding some UnitTests, I want to publish it to pypi. Is it ok for you if I publish it from my repository or do you want me to create a PR for your repo and you publish it to pypi?

    html = generate_html_output(nodedict, title, args.css)
  File "/usr2/kting/.local/lib/python3.7/site-packages/elf_size_analyze/html/gen.py", line 38, in generate_html_output
    overall_size = overall_size + y["cumulative_size"]
TypeError: unsupported operand type(s) for +: 'int' and 'NoneType'

Looks like gen is trying to access cumulative_size which is None if --no-cumulative-size is set

Dear @jedrzejboczar,

I could not find a better place to write this message, so I am opening it here and closing it right away to not mess up with your open issues 😄

Your script has been very instrumental in my analysis of embedded software binaries at work. To give you a small insight into the type of work we are doing, we had to strip the unnecessary code from our flight software binaries, and this is where your HTML visualization was especially useful. The workflow I was following was pretty simple: I printed the script's HTML output to PDF and went to highlight the most obvious candidates for removal from the flight executable along with the sizes that these components consume. After I had highlighted everything, I could quickly collect the candidate components and the total size that could be saved by removing all unused code. The HTML-to-PDF experience made the process very UX-friendly.

Thanks for putting this script together. It is definitely very useful.

Cheers,
Stanislav (Reflex Aerospace GmbH)

I tried running this but got the following error:

C:/msys64/clang64/bin/nm.exe: error: : unknown argument '--line-numbers'
Traceback (most recent call last):
  File "/home/david/.local/pipx/.cache/640b108cb1cbb91/bin/elf-size-analyze", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/home/david/.local/pipx/.cache/640b108cb1cbb91/lib/python3.11/site-packages/elf_size_analyze/__main__.py", line 72, in main
    fileinfo = extract_elf_symbols_fileinfo(args.elf, get_exe('nm'))
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/david/.local/pipx/.cache/640b108cb1cbb91/lib/python3.11/site-packages/elf_size_analyze/symbol.py", line 177, in extract_elf_symbols_fileinfo
    raise subprocess.CalledProcessError(nm_proc.returncode,
subprocess.CalledProcessError: Command '['nm', '--portability', '--line-numbers', '_build/projection-clang64-install/winrt/_winrt.pyd']' returned non-zero exit status 1.  

Apparently the LLVM version of nm does not support the --line-numbers option. llvm/llvm-project#39348

Is there a workaround?

Hi,

first of all thanks for the great alternative to size_report. But it seems that the reported size is not correct. For my elf file I have the following data points:

Linker output:

Memory region         Used Size  Region Size  %age Used
           FLASH:      920184 B     941568 B     97.73%
            SRAM:      235186 B       256 KB     89.72%
        IDT_LIST:          0 GB         2 KB      0.00%

The size_report --rom version reports me 920047 and your tool prints 816200. Unfortunately I cannot share the reports and don't really know how to help further to pin down the problem.

It seems that when generating HTML output for more than one section (e.g. elf-size-analyze -WFR) it is not possible to collapse a row. The javascript currently assumes there is only one generated document.

UnicodeDecodeError: 'gbk' codec can't decode byte 0xa6 in position 477: illegal multibyte sequence

it can be solved by (html/gen.py)
`
def generate_html_output(node_dict, title, custom_css=None):
table_content = ""

custom_css = custom_css or DEFAULT_CSS
with open(custom_css, encoding='utf-8') as f:
    css_styles = f.read()

with open(JAVASCRIPT, encoding='utf-8') as f:
    javascript = f.read()

... ...
`
open file with utf-8 encoding.

The size_report tool support so generate json reports. Would be great to see this here too to further process the reports.

which messes with display of symbols of templates and possibly other things.

Hi
i just noticed, that in some cases filepath case is not preserved and output is confusing.
For example i have both BatteryReader.cpp and batteryreader.hpp one section has some functions and other as the rest.
see this output:
out.txt

im using windows 10 machine and msp430-gcc in version 9.3.1

To disable WPS feature from WLAN_Lib_M2B_Standard_STA_AP, I tried to change #define WLAN_CONFIG_WPS 0 in wlanconfig.h. The first problem faced it was in wps_supplicant.c and wps_supplicant.h files. For example in wps_supplicant.h, functions's declarations and definitions depend on #ifdef CONFIG_WPS and #else. But in wps_supplicant.c same functions are defined without condition on ifdef CONFIG_WPS. To resolve this problem of redefinitions I tried to comment functions in wps_suplicant.c and I kept the functions defined in wps_supplicant.h that they return just 0 without doing nothing. The second problem faced was of #define IEEE8021X_EAPOL that is defined in wlan_cfg.h like that:
#if WLAN_CONFIG_WPS
#define CONFIG_WPS 1
#define EAP_WSC 1
#ifdef CONFIG_AP
#define EAP_SERVER_WSC 1
#endif /* CONFIG_AP /
#define IEEE8021X_EAPOL 1
#define EAP_TLS_NONE 1
#endif
As I disabled WLAN_CONFIG_WPS, we don't have a definition of "IEEE8021X_EAPOL". I tried to build and I had this error:
C:/git/STM32_Wifi/Firmware/Middlewares/ST/WLAN_Lib/wpa_supplicant/wpa_supplicant_bsd3/wpa_supplicant/wps_supplicant.c:452:7: error: 'struct wpa_ssid' has no member named 'eap'
When I looked in config_ssid.h, I found ;
#ifdef IEEE8021X_EAPOL
#define EAPOL_FLAG_REQUIRE_KEY_UNICAST BIT(0)
#define EAPOL_FLAG_REQUIRE_KEY_BROADCAST BIT(1)
/*
* eapol_flags - Bit field of IEEE 802.1X/EAPOL options (EAPOL_FLAG_*)
*/
int eapol_flags;

/**
 * eap - EAP peer configuration for this network
 */
struct eap_peer_config eap;

#endif /* IEEE8021X_EAPOL /
So the definition of eap struct depends on the definition of IEEE8021X_EAPOL in config_ssid.h but it doesn't depend on #ifdef IEEE8021X_EAPOL in wps_supplicant.c.
I need to understand this problem: why there are #ifdef IEEE8021X_EAPOL in file and not in other and also why we have #define IEEE8021X_EAPOL 1 inside of WPS enabling:
#if WLAN_CONFIG_WPS
#define CONFIG_WPS 1
#define EAP_WSC 1
#ifdef CONFIG_AP
#define EAP_SERVER_WSC 1
#endif / CONFIG_AP */
#define IEEE8021X_EAPOL 1
#define EAP_TLS_NONE 1
#endif

Could we have a possibility to fully disable WPS feature independently as a module?