unsaved changes: 0

[bol-van]

Whatever I change, save but not apply - subj is in the right-top corner. Changes can be seen in /tmp/.uci This bug lasts for years

[danielfdickinson]

@bol-van Can you confirm the version of openwrt and LuCI for which this occurs?

[danielfdickinson]

@bol-van Also what theme are you using?

[bol-van]

git-16.306.57050-d7d9772-1 theme - openwrt looks like bootstrap theme does not have this indication and thats why nobody cares because openwrt theme is abandoned

[jow-]

I cannot reproduce that issue with either Bootstrap or OpenWrt themes.

[danielfdickinson]

@bol-van Do you have any custom modifications? Out of memory? What device? Are you using a snapshot? From where? Did you build your own? From what repos, and what versions?

[bol-van]

Custom modifications : no, i didnt modify luci Out of memory : cant be. it happens even on 1gb ram device. What device : wndr3800, wndr3700, dir825b2, banana pi. All either run DD or LEDE. Own firmware : yes, i always build my own, never use snapshots. What repos : standard git checkout of latest DD/lede

[bol-van]

Experiment. I setup LEDE in vmware. Installed luci there. Works. Copied /usr/lib/lua/luci from problematic system to vmware lede. Replaced 3 .so inside to x64 version. Deleted /tmp/luci*. It still works ! Then I copied original luci from vmware lede to problematic system, replaced .so. It doesnt work. Looks like problem is not inside /usr/lib/lua/luci

[danielfdickinson]

That's some sort of compilation issue then (unclean tree?) Closing 'not a luci bug'

[bol-van]

I traced the reason to libuci.so. (file.c) uci_list_config_files enumerates all files in /etc/config including directories if directory is found then get_filename returns null configs[i] remains unset and in most cases contain NULL all part of code that use config enumeration stop on first NULL If a directory exist in /etc/config then all fails

    char **configs = NULL;
    char **p;
    if ((uci_list_configs(ctx, &configs) != UCI_OK) || !configs) {
        cli_perror();
        goto out;
    }

    for (p = configs; *p; p++) {
        package_cmd(cmd, *p);
    }

[bol-van]

This patch to file.c fixes the problem

838c838
<       int size, i;
---
>       int size, i, j;
862c862
<       for(i = 0; i < globbuf.gl_pathc; i++) {
---
>       for(i = 0, j = 0; i < globbuf.gl_pathc; i++) {
872c872
<               configs[i] = buf;
---
>               configs[j++] = buf;

[jow-]

Thanks for figuring this out @bol-van . This has been recently patched upstream in uci.git, see http://git.lede-project.org/49ec6ef .

[danielfdickinson]

Yes thank you @bol-van I obviously misdiagnosed; thanks @jow-