{
_bfd_error_handler
/* xgettext: c-format */
- (_("ERROR: %B is compiled for APCS-%d, whereas %B is compiled for APCS-%d"),
+ (_("error: %B is compiled for APCS-%d, whereas %B is compiled for APCS-%d"),
ibfd, obfd,
APCS_26_FLAG (ibfd) ? 26 : 32,
APCS_26_FLAG (obfd) ? 26 : 32
if (APCS_FLOAT_FLAG (ibfd))
/* xgettext: c-format */
- msg = _("ERROR: %B passes floats in float registers, whereas %B passes them in integer registers");
+ msg = _("error: %B passes floats in float registers, whereas %B passes them in integer registers");
else
/* xgettext: c-format */
- msg = _("ERROR: %B passes floats in integer registers, whereas %B passes them in float registers");
+ msg = _("error: %B passes floats in integer registers, whereas %B passes them in float registers");
_bfd_error_handler (msg, ibfd, obfd);
if (PIC_FLAG (ibfd))
/* xgettext: c-format */
- msg = _("ERROR: %B is compiled as position independent code, whereas target %B is absolute position");
+ msg = _("error: %B is compiled as position independent code, whereas target %B is absolute position");
else
/* xgettext: c-format */
- msg = _("ERROR: %B is compiled as absolute position code, whereas target %B is position independent");
+ msg = _("error: %B is compiled as absolute position code, whereas target %B is position independent");
_bfd_error_handler (msg, ibfd, obfd);
bfd_set_error (bfd_error_wrong_format);